G1

Source-backed capabilities

• Operating environments: Indoor grasp tests and outdoor okra field trial in Kagoshima, Staged human-robot sparring ring, Staged household-fixture demonstration, Temple of Heaven outdoor choreographed performance, Full-body avatar demonstration and data-collection setup, Pre-mapped WOW Summit conference floor, Staged demonstration at UFC Performance Institute, Televised staged robot kickboxing ring, Selected indoor acrobatics demonstration, Selected indoor dynamic-recovery demonstration, Selected physical dance demonstration, Physical sparse-stepping-stone research course, Indoor box-sorting research setup, Physical whole-body interaction research setup, Indoor physical motion-imitation research setup, Dual-robot collaborative manipulation research setup, Flat, gravel, grass and inclined physical test surfaces, controlled real-world humanoid dodgeball benchmark, Supermarket-replica research setup, Physical motion-tracking research setup, cluttered real-world indoor scenes, real-world humanoid loco-manipulation evaluations, Physical beam and box traversal research setup, Stationary tabletop manipulation research setup, Physical moving-target research trials, Lab and apartment research scenes with varied furniture layouts, humanoid-control research laboratory, Motion-capture-equipped physical research setup, Physical wall-barrier feasibility course, Qualitative physical trajectory-replay research setup, Shared-bin tabletop research setup, Physical active-perception manipulation research setup, Physical loaded-walking and stance research tests, Physical restricted-envelope arm-reaching setup, Indoor and outdoor soccer research tests, Physical whole-body manipulation research setup, Motion-capture-equipped long-rope research setup, Physical language-directed manipulation research setup, Outdoor natural-terrain research tests, Optical motion-capture-equipped research setup, Physical tabletop model-evaluation setup, mixed-terrain research course, interactive manipulation research laboratory, Physical whole-body teleoperation tests, Restricted tabletop bottle-grasping research workspace, Car-window glue-application research setup, Tabletop bimanual research setup, Indoor clutter, corridor, glass-wall and pedestrian research scenes, Humanoid demonstration-data collection setup, Stationary tabletop sensory-ablation research setup, Indoor treadmill and outdoor sidewalk research trials, Outdoor obstacle-avoidance research setup, Physical language-motion tracking research setup, Mostly static scene-aware whole-body research trials, OptiTrack-equipped object-transport research setup, Physical sparse-goal object-manipulation research setup, Pre-mapped dual-humanoid interaction research setup, Cadaveric endoscopic-surgery research operating room, Depth-perceived parkour research obstacle courses, Physical high-platform traversal research course, Indoor and outdoor skateboard research tests, Tabletop contact-control research trials, Physical text-to-motion research demonstrations, Staged indoor household-fixture research setup, Instrumented stair and compliant-ledge research course, Object-pose-tracked indoor loco-manipulation research setup, Dex3-1 tabletop manipulation research setup, Apartment and tabletop language-guided research setup, Physical locomotion research trials, Motion-capture-instrumented whole-body interaction research setup, Whole-body motion-tracking research tests, Language-conditioned tabletop robot-policy research setup, Tabletop fruit-placement data collection, Physical gait-control research tests, Full-body manipulation teleoperation research setup, Wired multi-camera learned-manipulation research setup, Indoor and outdoor posture-recovery research trials, Instrumented dynamic whole-body motion research trials, Staged home-task evaluation interiors, Shelving research setup, Tabletop towel-folding research setup, Tabletop phone-packaging research setup, Tabletop conveyor-sorting research setup, Tabletop plug-insertion research setup, Tabletop manipulation data-collection setup, Household-fixture teleoperation data-collection setup, FlashSAC physical G1 flat-ground and stair test course, Manufacturer-described configured manipulation applications
• Control modes: Autonomous, Teleoperation, Shared Control, Scripted, Planner Guided
• Auki conference guidance

  Demonstrated - 2025-09-09 - Guide visitors to selected conference booths using a pre-mapped Auki domain and onboard sensors. - environment: Pre-mapped WOW Summit conference floor

  Source
• Dynamic kip-up recovery

  Demonstrated - 2025-03-21 - Flip from lying on the back to a standing position in a kip-up demonstration. - environment: Selected indoor dynamic-recovery demonstration

  Source
• Perceptive stepping stones

  Research demonstration - 2026-10-13 - Cross four alternating stepping stones using chest-depth perception in a proof-of-concept trial. - environment: Physical sparse-stepping-stone research course

  Source
• Multi-terrain recovery

  Research demonstration - 2026-07-31 - Recover after a fall and resume commanded locomotion on flat ground, gravel, grass and slopes. - environment: Flat, gravel, grass and inclined physical test surfaces

  Source
• Dynamic locomotion

  Research demonstration - 2026-07-30 - Dodged different thrown balls using head-camera depth segmentation, succeeding on 95% of seeded real-world throws. - environment: controlled real-world humanoid dodgeball benchmark

  Source
• Navigation and obstacle avoidance

  Research demonstration - 2026-07-20 - Navigated cluttered indoor scenes without task-specific tuning after simulation-only training. - environment: cluttered real-world indoor scenes

  Source
• Reference-guided beam and box traversal

  Research demonstration - 2026-07-15 - Traverse a beam and step onto a box by tracking reference motions from EgoHTR. - environment: Physical beam and box traversal research setup

  Source
• Range-guided wall bypass

  Research demonstration - 2026-06-22 - Bypass wall barriers and recover toward a staged target without continuous joystick steering. - environment: Physical wall-barrier feasibility course

  Source
• Loaded walking with disturbance feedback

  Research demonstration - 2026-06-15 - Maintain commanded forward walking while carrying a one-sided hand load in disturbance tests. - environment: Physical loaded-walking and stance research tests

  Source
• Depth-guided natural-terrain traversal

  Research demonstration - 2026-06-03 - Traverse uneven ground, natural steps and slopes while adapting between walking and running. - environment: Outdoor natural-terrain research tests

  Source
• Multi-terrain locomotion

  Research demonstration - 2026-06-02 - Traversed boxes, hurdles, stairs and mixed terrain using onboard perception. - environment: mixed-terrain research course

  Source
• Visual-memory pedestrian and glass avoidance

  Research demonstration - 2026-04-01 - Navigate corridors and clutter, wait for doors to open, and avoid pedestrians or glass in supervised research trials. - environment: Indoor clutter, corridor, glass-wall and pedestrian research scenes

  Source
• Commanded running with onboard mini-PC

  Research demonstration - 2026-03-26 - Run on a treadmill or follow remotely set outdoor velocity and heading commands. - environment: Indoor treadmill and outdoor sidewalk research trials

  Source
• LiDAR-guided running avoidance

  Research demonstration - 2026-03-26 - Avoid detected outdoor obstacles while running under a LiDAR-based safety-planning layer. - environment: Outdoor obstacle-avoidance research setup

  Source
• Depth-guided parkour skill selection

  Research demonstration - 2026-02-17 - Vault obstacles, climb walls and land drops while a depth policy selects skills along commanded travel directions. - environment: Depth-perceived parkour research obstacle courses

  Source
• LiDAR-guided platform traversal

  Research demonstration - 2026-02-11 - Climb onto a platform, crawl across it and climb down through geometry-triggered whole-body skills. - environment: Physical high-platform traversal research course

  Source
• Skateboard propulsion and steering

  Research demonstration - 2026-02-03 - Push a skateboard, steer by leaning and transition between pushing and coasting phases. - environment: Indoor and outdoor skateboard research tests

  Source
• Dual-depth stair ascent and descent

  Research demonstration - 2025-10-16 - Climb and descend a three-step flight using fused chest and hip depth views. - environment: Instrumented stair and compliant-ledge research course

  Source
• Compliant ledge traversal

  Research demonstration - 2025-10-16 - Step across a hinged plank and onto a compliant mat at a 36cm ledge. - environment: Instrumented stair and compliant-ledge research course

  Source
• Commanded walk-run transitions

  Research demonstration - 2025-09-25 - Transition smoothly between walking and running while following velocity commands up to 2.5 metres per second. - environment: Physical locomotion research trials

  Source
• Gait-conditioned walking and standing

  Research demonstration - 2025-05-27 - Walk, stand and transition from walking to standing under supplied motion commands. - environment: Physical gait-control research tests

  Source
• Proprioceptive standing-up recovery

  Research demonstration - 2025-02-12 - Rise from lying, seated or leaning postures across indoor surfaces and outdoor grass, stone and platforms. - environment: Indoor and outdoor posture-recovery research trials

  Source
• Flashsac blind walking

  Research demonstration - Walk forward, backward and sideways using a proprioceptive locomotion policy. - environment: FlashSAC physical G1 flat-ground and stair test course

  Source
• Flashsac blind stairs

  Research demonstration - Climb stairs with 15-centimetre risers using a blind locomotion policy. - environment: FlashSAC physical G1 flat-ground and stair test course

  Source
• Configured okra grasping

  Demonstrated - 2026-09-29 - Detect, reach and grasp okra with a camera-guided right-hand cutter attachment; cutting remained unfinished. - environment: Indoor grasp tests and outdoor okra field trial in Kagoshima

  Source
• Requested package retrieval

  Demonstrated - 2026-07-20 - Retrieve a requested coloured medicine package from stacked boxes. - environment: Staged household-fixture demonstration

  Source
• Two-controller loco-manipulation

  Research demonstration - 2026-10-05 - Wipe a sink while crouched, move carts or chairs, and pick up a low box while an operator supplies two hand targets. - environment: Physical whole-body interaction research setup

  Source
• Reference-tracked suitcase pulling

  Research demonstration - 2026-10-05 - Pull a suitcase while walking along the supplied motion reference. - environment: Indoor physical motion-imitation research setup

  Source
• Reference-tracked tripod carrying

  Research demonstration - 2026-10-05 - Grasp and carry a tripod using Inspire hands synchronized with the body reference. - environment: Indoor physical motion-imitation research setup

  Source
• Dual-robot handling with Vega1

  Research demonstration - 2026-09-22 - Clear or nudge objects, tilt a board and receive a doll while coordinating with a Vega1 partner. - environment: Dual-robot collaborative manipulation research setup

  Source
• Chip-bag shelf placement

  Research demonstration - 2026-07-22 - Grasp a chip bag from a trolley box, rotate toward a shelf and place it using an offboard learned policy. - environment: Supermarket-replica research setup

  Source
• Whole-body loco-manipulation

  Research demonstration - 2026-07-20 - Completed 71 of 80 trials across eight real-world loco-manipulation task families using persistent 3D object-state verification. - environment: real-world humanoid loco-manipulation evaluations

  Source
• Stationary pear placement

  Research demonstration - 2026-06-30 - Pick a pear from the table and place it into a basket. - environment: Stationary tabletop manipulation research setup

  Source
• Learned page turning

  Research demonstration - 2026-06-30 - Lift and turn a page until it lies flat on the other side. - environment: Stationary tabletop manipulation research setup

  Source
• Bimanual book insertion

  Research demonstration - 2026-06-30 - Hold a document bag with one hand and insert a book with the other. - environment: Stationary tabletop manipulation research setup

  Source
• Learned bottle-cap unscrewing

  Research demonstration - 2026-06-30 - Hold a bottle and rotate its cap to unscrew it. - environment: Stationary tabletop manipulation research setup

  Source
• Externally tracked moving-target reaching

  Research demonstration - 2026-06-29 - Replan whole-body movement to reach a moving target whose pose is supplied by an external tracker. - environment: Physical moving-target research trials

  Source
• Vision-guided box handling

  Research demonstration - 2026-06-29 - Approach selected objects, turn around them, and pick or place boxes on the floor and raised surfaces. - environment: Lab and apartment research scenes with varied furniture layouts

  Source
• Trajectory tracking

  Research demonstration - 2026-06-24 - Tracked asynchronous upper-body task-space trajectories on the physical humanoid. - environment: humanoid-control research laboratory

  Source
• Motion-capture-guided ball contact

  Research demonstration - 2026-06-24 - Hit forehand and backhand tennis shots and contact a soccer ball using externally tracked targets. - environment: Motion-capture-equipped physical research setup

  Source
• Motion-capture-guided box handling

  Research demonstration - 2026-06-24 - Pick up and place a box using TaskNPoint targets and external motion capture. - environment: Motion-capture-equipped physical research setup

  Source
• Dex3 recorded-trajectory replay

  Research demonstration - 2026-06-22 - Replay walking-and-grasping motions and pull a mock refrigerator door while stepping backward. - environment: Qualitative physical trajectory-replay research setup

  Source
• Language-conditioned bin placement

  Research demonstration - 2026-06-19 - Select a red ball, cup, bottle or medicine bottle from language input and place it in a shared bin. - environment: Shared-bin tabletop research setup

  Source
• Active-neck grasp and transfer

  Research demonstration - 2026-06-17 - Pick and transfer bread, bags or towels, including grasping while crouched, using an active camera neck. - environment: Physical active-perception manipulation research setup

  Source
• Restricted-envelope arm reaching

  Research demonstration - 2026-06-14 - Move the left arm to sampled reachable Cartesian targets using an energy-aware policy. - environment: Physical restricted-envelope arm-reaching setup

  Source
• Camera-guided whole-body handling

  Research demonstration - 2026-06-08 - Push and load a cart, raise a basket, retrieve a bag and stock objects on shelves through whole-body motion. - environment: Physical whole-body manipulation research setup

  Source
• VLM-directed whole-body task sequences

  Research demonstration - 2026-06-04 - Pick, carry and place objects, including crouched and two-arm sequences, using the tested perception-and-planning stack. - environment: Physical language-directed manipulation research setup

  Source
• Dex1 world-model manipulation

  Research demonstration - 2026-06-03 - Open a pot lid and place a potato inside, select a red bottle, or place an object on a marked target. - environment: Physical tabletop model-evaluation setup

  Source
• Object manipulation

  Research demonstration - 2026-05-29 - Grasp gaze-selected objects and hand them to a person using voice-triggered human target selection. - environment: interactive manipulation research laboratory

  Source
• Pose-tracked bottle grasping

  Research demonstration - 2026-04-19 - Track and grasp a drink bottle at five desk positions, then lift it with a Dex3 hand. - environment: Restricted tabletop bottle-grasping research workspace

  Source
• XL-VLA bimanual tasks

  Research demonstration - 2026-04-15 - Prepare fruit, pass a bottle between hands, and pour sauce or sugar using the XL-VLA policy with Inspire hands. - environment: Tabletop bimanual research setup

  Source
• Active-camera cube placement

  Research demonstration - 2026-03-30 - Find small cubes placed across the workspace and grasp and place them into a box. - environment: Stationary tabletop sensory-ablation research setup

  Source
• Image-conditioned static-scene actions

  Research demonstration - 2026-03-10 - Move a box, sit on a sofa or chair, or kick a ball after generating a motion plan from the initial view. - environment: Mostly static scene-aware whole-body research trials

  Source
• MoCap-guided object-motion tracking

  Research demonstration - 2026-03-03 - Lift or carry boxes and transport a suitcase while tracking dense motion references with external object poses. - environment: OptiTrack-equipped object-transport research setup

  Source
• Depth-guided operator goal following

  Research demonstration - 2026-03-03 - Move boxes toward operator-specified future poses using onboard depth observations and the learned controller. - environment: Physical sparse-goal object-manipulation research setup

  Source
• Compliant surface contact

  Research demonstration - 2026-01-21 - Follow a curved surface and press a held massage ball against a tabletop object. - environment: Tabletop contact-control research trials

  Source
• Asymmetric egg and bottle placement

  Research demonstration - 2026-01-21 - Place an egg and a sauce bottle with separate compliance settings for each arm. - environment: Tabletop contact-control research trials

  Source
• Fork contact and object lifting

  Research demonstration - 2026-01-21 - Poke produce with a held fork and grasp and lift tabletop objects. - environment: Tabletop contact-control research trials

  Source
• GR00T household object retrieval

  Research demonstration - 2025-12-15 - Approach household fixtures to rack a mug, retrieve a snack or eraser, box a shoe and place an apple in a basket. - environment: Staged indoor household-fixture research setup

  Source
• Rubber-hand cube contact and transfer

  Research demonstration - 2025-10-13 - Approach a cube to press or push it, and transfer a cube between hands. - environment: Object-pose-tracked indoor loco-manipulation research setup

  Source
• Parallel-gripper cube pickup

  Research demonstration - 2025-10-13 - Approach a cube, grasp it with the custom parallel grippers and lift it. - environment: Object-pose-tracked indoor loco-manipulation research setup

  Source
• Parallel-gripper cabinet opening

  Research demonstration - 2025-10-13 - Reach for a cabinet handle and pull it open with a custom gripper. - environment: Object-pose-tracked indoor loco-manipulation research setup

  Source
• Learned chair rotation

  Research demonstration - 2025-10-09 - Rotate a chair with the configured Dex3-1 hands. - environment: Dex3-1 tabletop manipulation research setup

  Source
• Learned tabletop towel folding

  Research demonstration - 2025-10-09 - Fold a towel laid on a desk. - environment: Dex3-1 tabletop manipulation research setup

  Source
• Learned toy placement and handover

  Research demonstration - 2025-10-09 - Place a dumpling toy into a plate or hand it to a person. - environment: Dex3-1 tabletop manipulation research setup

  Source
• Language-commanded tabletop arrangement

  Research demonstration - 2025-10-06 - Pick up a bottle or toy and arrange fruit and toys relative to other tabletop objects. - environment: Apartment and tabletop language-guided research setup

  Source
• Language-commanded bottle relocation

  Research demonstration - 2025-10-06 - Carry a Coke bottle through the apartment and place it on the requested counter. - environment: Apartment and tabletop language-guided research setup

  Source
• Motion-capture-guided door traversal

  Research demonstration - 2025-09-20 - Push a door open by hand, walk through and return by kicking it open. - environment: Motion-capture-instrumented whole-body interaction research setup

  Source
• Language-selected object placement

  Research demonstration - 2025-06-11 - Place the requested fruit, soap or snack packet onto a plate among tabletop distractors. - environment: Language-conditioned tabletop robot-policy research setup

  Source
• Teleoperated fruit placement

  Research demonstration - 2025-06-01 - Pick apples, pears, starfruit or grapes and place them on a plate under upper-body teleoperation. - environment: Tabletop fruit-placement data collection

  Source
• Wired learned tomato handling

  Research demonstration - 2025-02-18 - Squat to pick a tomato from a low sofa and pick and place a tomato using a learned policy. - environment: Wired multi-camera learned-manipulation research setup

  Source
• Model-controlled towel folding

  Research demonstration - Fold a towel on a tabletop using the evaluation’s dexterous hand configuration. - environment: Tabletop towel-folding research setup

  Source
• Model-controlled plug insertion

  Research demonstration - Insert a plug into a tabletop power strip in the configured evaluation. - environment: Tabletop plug-insertion research setup

  Source
• Inspire teleoperated washer loading

  Research demonstration - Lift clothing from a pile and load it into a front-loading washing machine under human control. - environment: Household-fixture teleoperation data-collection setup

  Source
• Dex3-1 intended object handling

  Claimed - 2024-05-16 - Handle eggs, open soda bottles and manipulate hard objects with the optional Dex3-1 hands. - environment: Manufacturer-described configured manipulation applications

  Source
• Fifty-robot martial-arts formation

  Demonstrated - 2026-02-23 - Perform synchronized leaps, shifts, flips and a fist-clasping salute in a 50-robot formation. - environment: Temple of Heaven outdoor choreographed performance

  Source
• Dual-G1 contact coordination

  Research demonstration - 2026-03-03 - Synchronize hugs, greetings and contact routines with a second G1 using shared pose and timing information. - environment: Pre-mapped dual-humanoid interaction research setup

  Source
• Language-commanded apartment navigation

  Research demonstration - 2025-10-06 - Approach a named sofa, fridge or television and stop near the target. - environment: Apartment and tabletop language-guided research setup

  Source
• World-model-controlled sparring

  Demonstrated - 2026-09-07 - Step, punch and kick while sparring with a human in a staged ring demonstration. - environment: Staged human-robot sparring ring

  Source
• Motion-capture avatar teleoperation

  Demonstrated - 2025-11-05 - Mirror a motion-capture-suit operator to kick a ball, box and perform household tidying motions. - environment: Full-body avatar demonstration and data-collection setup

  Source
• Staged athlete fighting interaction

  Demonstrated - 2025-08-28 - Perform fighting motions while interacting with athletes at the UFC Performance Institute. - environment: Staged demonstration at UFC Performance Institute

  Source
• Robot-versus-robot kickboxing

  Demonstrated - 2025-05-31 - Punch and kick another G1 in a staged boxing ring. - environment: Televised staged robot kickboxing ring

  Source
• Reference motion replay

  Research demonstration - 2026-07-22 - Replay locomotion, crouching and dancing references while maintaining balance during manual pushes. - environment: Physical motion-tracking research setup

  Source
• Optical motion-capture teleoperation

  Research demonstration - 2026-06-03 - Track an operator’s dense body references or sparse hand targets derived from optical motion capture. - environment: Optical motion-capture-equipped research setup

  Source
• IMU whole-body teleoperation

  Research demonstration - 2026-05-12 - Follow an operator’s walking, sit-stand, turning, bowing and arm gestures through IMU retargeting. - environment: Physical whole-body teleoperation tests

  Source
• Dex3 whole-body teleoperation

  Research demonstration - 2026-03-31 - Retarget an operator’s finger, wrist and body inputs to dexterous manipulation and locomotion commands. - environment: Humanoid demonstration-data collection setup

  Source
• Externally generated language motions

  Research demonstration - 2026-03-17 - Execute walking, squatting, kicking or hand-gesture references generated from language on an external server. - environment: Physical language-motion tracking research setup

  Source
• Teleoperated endoscope positioning

  Research demonstration - 2026-02-27 - Position a rigid endoscope under continuous VR control while a surgeon performs a cadaver procedure. - environment: Cadaveric endoscopic-surgery research operating room

  Source
• Language-generated whole-body pantomime

  Research demonstration - 2026-01-19 - Mime an elephant or violin playing, and take steps before miming a phone call. - environment: Physical text-to-motion research demonstrations

  Source
• Assisted staircase gesture sequence

  Research demonstration - 2025-09-20 - Climb stairs, bow, sit, wave and jump down in a sequence with occasional human support. - environment: Motion-capture-instrumented whole-body interaction research setup

  Source
• Locked-wrist whole-body motion tracking

  Research demonstration - 2025-07-10 - Track supplied whole-body gestures, steps and dance-like motion sequences with the wrist joints locked. - environment: Whole-body motion-tracking research tests

  Source
• Exoskeleton-controlled handling

  Research demonstration - 2025-02-18 - Move objects between low and high shelves, open an oven and operate a handheld barcode scanner under cockpit control. - environment: Full-body manipulation teleoperation research setup

  Source
• ASAP dynamic motion tracking

  Research demonstration - 2025-02-03 - Track kicking, forward-jump, single-foot balance and celebration-motion references. - environment: Instrumented dynamic whole-body motion research trials

  Source
• Standing side flip

  Demonstrated - 2025-04-02 - Launch from standing, rotate sideways in the air and land back on the feet. - environment: Selected indoor acrobatics demonstration

  Source
• Westlake-model ballet sequence

  Demonstrated - 2025-03-21 - Turn on one leg, perform two airborne kicks and finish with a stable bow in a ballet demonstration. - environment: Selected physical dance demonstration

  Source
• Perception-guided soccer shooting

  Research demonstration - 2026-06-09 - Aim and kick a soccer ball using onboard perception and a marked target. - environment: Indoor and outdoor soccer research tests

  Source
• Camera-guided soccer kick

  Research demonstration - 2026-06-08 - Kick a soccer ball into a goal net using the MotionWAM whole-body policy. - environment: Physical whole-body manipulation research setup

  Source
• Motion-capture-guided rope skipping

  Research demonstration - 2026-06-06 - Coordinate rope turning with another humanoid or human, including single human and humanoid jumpers. - environment: Motion-capture-equipped long-rope research setup

  Source
• Omni-modal pillow and clothing tidying

  Demonstrated - 2026-07-20 - Pick up a pillow for the sofa and sort clothing into a laundry basket. - environment: Staged household-fixture demonstration

  Source
• Omni-modal dishwasher loading

  Demonstrated - 2026-07-20 - Load a plate into a dishwasher in a staged household demonstration. - environment: Staged household-fixture demonstration

  Source
• Material handling

  Research demonstration - 2026-10-06 - Place boxes on shelves by hand and with a kick, recovering after pushes or box removal in Workhorse evaluations. - environment: Indoor box-sorting research setup

  Source
• Reference-controlled box lift

  Research demonstration - 2026-07-22 - Stand from a crouch while lifting a pre-held 1.5-kilogram box through a reference motion. - environment: Physical motion-tracking research setup

  Source
• Camera-guided whiteboard wiping

  Research demonstration - 2026-06-08 - Wipe a whiteboard while coordinating leg, torso and arm motion. - environment: Physical whole-body manipulation research setup

  Source
• Camera-guided laundry and waste placement

  Research demonstration - 2026-06-08 - Place clothes into a washing machine and throw waste into a bin in configured research tasks. - environment: Physical whole-body manipulation research setup

  Source
• Programmed window-glue application

  Research demonstration - 2026-04-19 - Move a glue stick along a car quarter-window edge using a configured trajectory template. - environment: Car-window glue-application research setup

  Source
• Active-camera can sorting

  Research demonstration - 2026-03-30 - Grasp beverage cans at known locations and sort them by colour using a configured ACT policy. - environment: Stationary tabletop sensory-ablation research setup

  Source
• Rubber-hand box lifting

  Research demonstration - 2025-10-13 - Walk to a box and lift it from the floor using the configured rubber hands. - environment: Object-pose-tracked indoor loco-manipulation research setup

  Source
• Parallel-gripper board erasing

  Research demonstration - 2025-10-13 - Hold an eraser in a custom gripper and wipe a board. - environment: Object-pose-tracked indoor loco-manipulation research setup

  Source
• Learned desk erasing

  Research demonstration - 2025-10-09 - Move a held eraser across the desk surface. - environment: Dex3-1 tabletop manipulation research setup

  Source
• Motion-capture-guided load relocation

  Research demonstration - 2025-09-20 - Kneel, lift and relocate a suitcase, box or stack of foam mats. - environment: Motion-capture-instrumented whole-body interaction research setup

  Source
• Model-controlled home tidying

  Research demonstration - Move clothes toward a washing machine, bag household waste and tidy kitchen items during model evaluations. - environment: Staged home-task evaluation interiors

  Source
• Model-controlled shelf placement

  Research demonstration - Carry objects and place them on shelves in the manufacturer’s whole-body evaluation. - environment: Shelving research setup

  Source
• Model-controlled phone packaging

  Research demonstration - Arrange a phone and its packaging in a tabletop model evaluation. - environment: Tabletop phone-packaging research setup

  Source
• Model-controlled conveyor sorting

  Research demonstration - Pick building blocks from a moving conveyor and place them in tabletop bins. - environment: Tabletop conveyor-sorting research setup

  Source
• Dex1 tabletop wiping

  Research demonstration - Wipe material across a tabletop with a cloth held in a Dex1 gripper. - environment: Tabletop manipulation data-collection setup

  Source
• Dex1 motor-part assembly recording

  Research demonstration - Grasp and align motor parts at a tabletop assembly station. - environment: Tabletop manipulation data-collection setup

  Source
• Brainco teleoperated dish loading

  Research demonstration - Pick a plate beside the sink and place it into an open dishwasher under human control. - environment: Household-fixture teleoperation data-collection setup

  Source
• Intended configured welding

  Claimed - 2024-05-16 - Perform welding operations with a task-equipped G1 EDU configuration. - environment: Manufacturer-described configured manipulation applications

  Source

• Reported Output

  Roughly 10,000–20,000 G1 robots produced cumulatively - 2026-10-22 - Wang Xingxing disclosed a cumulative G1 production range at the World Physical AI Conference.

  Source

Exact-product configurations are shown without inferring supply, purchase, customer status, standard fit, deployment or partnership.

• Orboh okra-field manipulation stack

  2026-09-29 - Research Experiment - Tested With - Physical - Orboh tested a customised Unitree G1 that detected, approached and grasped okra in a real commercial field.

  Source
• MiraiBit Physical-AI R&D environment

  2026-09-07 - Research Experiment - Tested With - Physical - MiraiBit added a physical Unitree G1 to its development environment for humanoid Physical-AI experiments.

  Source
• MANUS gloves

  2026-04-15 - Research Experiment - Controlled By - Physical - XL-VLA research used MANUS gloves in a HOMIE-derived pipeline to control a Unitree G1's Inspire Hand across four bimanual tasks.

  Source
• MANUS gloves

  2026-03-31 - Research Experiment - Controlled By - Physical - USC PSI Ψ₀ research used MANUS gloves plus PICO wrist/head tracking to teleoperate a Unitree G1 fitted with Dex3-1 hands.

  Source
• MANUS data gloves

  2026-02-18 - Digital Simulation - Tested With - Simulated - TeleOpBench evaluates a simulated Unitree G1 using MANUS finger capture with Xsens body tracking. The paper reports separate physical dual-arm validation without naming that hardware as G1.

  Source
• MANUS Quantum Metagloves

  2026-01-13 - Research Experiment - Controlled By - Physical - METIS research, previously reported as RoboBrain-Dex, used MANUS Quantum Metagloves with a Unitree G1 and Inspire hands for egocentric dexterity-data collection.

  Source
• Imacreate humanoid motion-reproduction software (formal name unannounced)

  Physical Demo - Controlled By - Physical - Imacreate’s software controls a Unitree G1 in kendama and diabolo demonstrations using recorded human actions and residual correction.

  Source
• Inspire Hand (exact model undisclosed)

  Physical Demo - Mounted - Physical - GenON demonstrated its GenP care-assistance software on a Unitree G1 fitted with Inspire hands at AI EXPO KOREA in May 2026.

  Source
• Dexterous hands (exact model unspecified)

  Research Experiment - Mounted - Physical - KinoGen’s physical Unitree G1 experiments use dexterous hands whose exact model is not specified.

  Source
• Acemate ball-feeding machine

  Research Experiment - Tested With - Physical - AdaPT tests physical G1 tennis rallies with an Acemate ball-feeding machine producing repeatable launch trajectories.

  Source
• Unitree Dex-3 hand

  Research Experiment - Mounted - Physical - AdaPT tests a mounted Unitree Dex-3 hand on G1 for reduced-swing amateur tennis serving, separately from its professional-style gripper setup.

  Source
• Synria Gloria-D parallel gripper

  Research Experiment - Mounted - Physical - AdaPT’s G1 serving evaluation uses a mounted Synria Gloria-D parallel gripper to grasp and toss the tennis ball.

  Source
• Noitom MCC-400 motion-capture cameras

  Research Experiment - Tested With - Physical - AdaPT’s main physical G1 tennis experiments use 35 Noitom MCC-400 cameras to track robot-root and ball positions.

  Source
• OCLO controller

  Research Experiment - Controlled By - Physical - Unitree G1 performs loco-manipulation tasks with OCLO posture and compliance control; an operator commands hand targets through two controllers.

  Source
• S1

  Research Experiment - Controlled By - Mixed - A Unitree G1 played a physical football match using a Skild S1-derived policy post-trained through simulated self-play.

  Source

Supply / Components

• G1

  Embedded Compute - Compute & Control - NVIDIA - Unitree G1-EDU incorporates an NVIDIA Jetson Orin NX development computer - source: docs.westonrobot.com

  Source
• Unitree G1

  Mid 360 LIDAR For Unitree G1 - sensing - Livox - Unitree G1 incorporates a Livox MID-360 lidar in its head - source: support.unitree.com

  Source
• G1

  Realsense D435i Depth Camera For G1 - Perception - RealSense - Unitree G1 incorporates a RealSense D435i depth camera in its head - source: github.com

  Source
• MID-360

  sensing

• VLM-Guided Contact Retargeting and Generative Partner Modeling for Humanoid-Human Interaction

  2026-10-20 - Research Publication - REACT transfers semantic contact structure from human demonstrations to humanoid interaction policies. A physical Unitree G1 performs high-fives, hugs, handshakes, pats on the back and linked-arm interactions using onboard sensing alone.

  Source
• CusADi2: Latency-Optimized Batch Computation of Robot Kinematics, Jacobians, and Hessians on the GPU

  2026-10-18 - Research Publication - CusADi2 reduces batch-computation latency by parallelizing both across and within robot-model evaluations on GPUs. The authors benchmark kinematics, Jacobians and Hessians across67 MuJoCo Menagerie robots and report lower latency at batch sizes64–4096 than evaluated alternatives.

  Source
• Vision as Proprioception: Estimating Humanoid Robot Pose from Video

  2026-10-18 - Research Publication - HumanoidPose estimates humanoid joint configurations and root pose from ordinary video using pretrained human-vision features and a kinematics-conditioned decoder. The ETH Zürich team trains across16 synthetic embodiments and reports18–32% lower joint-angle error than its strongest evaluated baselines on three real robot platforms.

  Source
• Whole-Body Dexterity

  2026-10-15 - Research Publication - Whole-Body Dexterity learns to reposition objects using a humanoid’s arms and torso as contact surfaces. A simulated grasp-generation and policy-distillation pipeline transfers to physical Unitree G1 manipulation with coarse human motion guidance.

  Source
• Compliant GMT: Making General Motion Tracking Compliant via Residual Reference Adaptation

  2026-10-15 - Research Publication - Compliant GMT adds a residual reference adapter to a pretrained humanoid motion tracker. Simulation and physical Unitree G1 experiments investigate whole-body compliance, contact response and robustness to disturbances.

  Source
• CQDC-Warp: GPU-Accelerated Differentiable Quasi-Dynamic Simulation for Contact-Rich Manipulation

  2026-10-15 - Research Publication - CQDC-Warp uses GPU-batched differentiable quasi-dynamic simulation to plan contact-rich trajectories. Experiments cover dexterous-hand and whole-body models, with a limited physical replay on a Sharpa hand.

  Source
• PingpongDuo: Agile Humanoid Table Tennis Doubles with Safe Team Coordination

  2026-10-14 - Research Publication - PingpongDuo combines team-level risk-aware positioning with separate footwork and striking policies for humanoid table-tennis doubles. Physical experiments use two 29-DoF Unitree G1 robots with external motion capture and demonstrate more than ten consecutive alternating strikes.

  Source
• KinoGen: Customizable Humanoid Loco-Manipulation References via Kinematically Grounded Video Generation

  2026-10-14 - Research Publication - KinoGen generates robot-specific humanoid motion references by conditioning video generation on calibrated renders and fitting the robot and objects back to the two-view output. Tracking policies transfer selected tasks to a physical Unitree G1.

  Source
• PanTact: Teaching Whole-Body Dexterity with a Wearable Contact Interface

  2026-10-08 - Research Publication - PanTact learns humanoid whole-body manipulation from robot-free human demonstrations collected with contact patches, egocentric vision and bimanual UMI grippers. A policy predicts contact goals and a whole-body controller executes them on Unitree G 1 without added body contact sensors. Research demonstrations include grasp-and-embrace, sweep-and-stow, carry-and-push and load-sharing.

  Source
• Being-M0.7: A Latent World-Action Model for Humanoid Robots

  2026-10-08 - Research Publication - Being-M0.7 transfers visual-motion priors from over 10,000 hours of raw human-centric data to humanoid control through pre-training, robot mid-training and action post-training. The study reports 71.1% success over 180 SIMPLE simulation trials and 13/15 successes on three physical Unitree G1 tasks, tying GR00T-N1.6.

  Source
• Workhorse: Learning Robust Whole-Body Humanoid Loco-Manipulation from Human Data

  2026-10-06 - Research Publication - Workhorse learns autonomous humanoid locomotion and manipulation from robot-free human demonstrations, combining a visual planner with a reinforcement-learning tracker.

  Source
• Dataset-Free Compliant Humanoid Loco-Manipulation with Dynamic Online Posture

  2026-10-05 - Research Publication - OCLO learns compliant humanoid locomotion and manipulation without human-motion datasets. On Unitree G1, two hand controllers direct end-effector targets while the system adjusts posture and balance under external forces.

  Source
• InterMimicGen: Scaling Humanoid Loco-Manipulation through Self-Evolving Motion Imitation

  2026-10-05 - Research Publication - InterMimicGen couples humanoid motion references with a learned tracker in an iterative data-generation loop. It retargets human-object interactions to several embodiments and demonstrates selected motions on physical G1, Booster K1 and Dexmate Vega-1 Pro robots.

  Source
• Physical Self-Play

  2026-09-23 - Research Publication - A scoring-only self-play objective produced dribbling, shielding, tackling and recovery behaviours after 140 simulated years of play in NVIDIA Isaac Sim.

  Source
• ViLoMan: Learning Visual-Proprioceptive Whole-Body Loco-Manipulation Skills for Humanoid Robots

  2026-09-16 - Research Publication - A Unitree G1 closes doors using onboard depth and proprioception without reference motion or intermediate commands at deployment.

  Source
• Weave: Learning Whole-Body Dexterous Loco-Manipulation from Human-Object Interactions

  2026-09-15 - Research Publication - Weave retargets human-object interactions to train a contact-aware whole-body humanoid policy controlling 29 body and 12 finger joints. It reports 92.5% success on trained interactions and 65% on unseen sequences.

  Source
• UnifoLM-WLA-1.0: A Generalist Robot Policy for Whole-body Interactive Autonomy

  2026-09-10 - Research Publication - Unitree presents UnifoLM-WLA-1.0 as a 6-billion-parameter embodied foundation model trained on approximately 2,500 hours of real-robot data. The company reports one model covering 54 tabletop tasks and 10 whole-body mobile-manipulation tasks on Unitree G1, with support for two-finger grippers and multiple five-finger dexterous hands.

  Source
• Towards Professional Tennis Styles for Humanoid Robots with Adaptive Motion Planning and Tracking

  2026-08-20 - Research Publication - AdaPT learns tennis rallying and serving styles from video and motion capture, adapting motion timing to reduce tracking errors. Physical experiments use Unitree G1 and Dobot Atom P3; the public code release covers first-stage G1 adaptive serve tracking with a checkpoint and two motion clips.

  Source
• HiPHI: A Large-Scale Benchmark for High-Precision Human Motion and Object-Interaction

  2026-08-17 - Research Publication - A large public motion and interaction dataset evaluated on Unitree G1.

  Source
• GigaBrain-0.7: Scaling Embodied Foundation Models to Emergent Capabilities with a Three-System Architecture

  2026-08-16 - Research Publication - GigaBrain-0.7 combines action control, understanding and planning, and world-model prediction in a three-system embodied model. GigaAI reports training on over 37,000 hours of heterogeneous data and physical evaluation on Maker H01 and AgileX PiPER-family manipulators.

  Source
• PAC-MAN: Perception-Aware CBF-RL for Whole-Body Safety in Humanoid Dodgeball

  2026-07-30 - Research Publication - A physical Unitree G1 dodged seeded ball throws using onboard perception, with 95% success in the reported benchmark.

  Source
• Learning Reusable Hybrid Motion Priors for Humanoid Locomotion from Motion Imitation

  2026-07-27 - Research Publication - A zero-shot G1 deployment walked, sidestepped and turned while tracking commanded velocity.

  Source
• Closing the Lab-to-Store Gap: A Data-Efficient Post-Training and Experience-Driven Learning VLA Framework for Retail Humanoids

  2026-07-22 - Research Publication - A G1-Edu using GR00T N1.6 was evaluated on real supermarket chip restocking after data-efficient post-training.

  Source
• What Matters in Humanoid General Motion Tracking? An Empirical Study

  2026-07-22 - Research Publication - Physical G1 tests demonstrate diverse motion tracking, perturbation resistance and forceful interaction.

  Source
• Learning Adaptive Safety Margins for Visual Navigation

  2026-07-20 - Research Publication - A simulation-trained safety-margin selector transferred to G1 and navigated cluttered indoor scenes without task-specific tuning.

  Source
• Closing the Loop in Humanoid VLA: Persistent 3D Object Tokens for Verifiable Loco-Manipulation

  2026-07-20 - Research Publication - POT-VLA reached 71 of 80 successes across eight real-world Unitree G1 task families.

  Source
• Semantic Audio-driven Understanding for Dynamic Humanoid Whole Body Control

  2026-07-15 - Research Publication - A physical G1 followed music-selected whole-body policy sequences in a live validation.

  Source
• EgoHTR: Egocentric 4D Demonstrations of Human Terrain Traversal

  2026-07-15 - Research Publication - EgoHTR-trained policies reproduce beam and box-up terrain motions on a physical G1.

  Source
• zengury/roboonto

  2026-07-02 - Research Publication - Machine-readable robot ontologies for LLM agents — query what a robot can do, validate actions before they touch hardware. IEEE 1872-anchored. Ships Unitree G1 & AgiBot X2 packs.

  Source
• RoboTacDex: A Dexterous Visual-Tactile-Action Dataset for Humanoid Manipulation

  2026-06-30 - Research Publication - In the field of robot learning, large-scale and diverse demonstration trajectories provide the fundamental basis for enhancing robotic manipulation ability.

  Source
• VLK: Learning Humanoid Loco-Manipulation from Synthetic Interactions in Reconstructed Scenes

  2026-06-29 - Research Publication - Perception-based humanoid loco-manipulation requires connecting egocentric observations and task instructions to whole-body motion.

  Source
• ReactiveBFM: Reactive Closed-Loop Motion Planning Towards Universal Humanoid Whole-Body Control

  2026-06-29 - Research Publication - While current Behavior Foundation Models (BFMs) provide robust control priors for humanoids, they only execute pre-defined reference motions.

  Source
• TaskNPoint: How to Teach Your Humanoid to Hit a Backhand in Minutes

  2026-06-24 - Research Publication - How do we learn to hit a tennis backhand?

  Source
• Learning Asynchronous Upper-body Task-space Trajectory Tracking Policy for Humanoid Robots

  2026-06-24 - Research Publication - High-level humanoid planners often output sparse task-space, low-rate trajectories, whereas whole-body controllers run at high frequency.

  Source
• RGB: RL Guided Whole-Body MPPI for Humanoid Control

  2026-06-23 - Research Publication - Humanoid robots require whole-body controllers that are both robust and precise in contact-rich environments.

  Source
• TEXEDO : Test Time Scaling for Controller-aware Language-conditioned Humanoid Motion Generation

  2026-06-22 - Research Publication - Text-conditioned motion generation is a promising interface for programming humanoid robots, yet current generators are often trained on human motion datasets retargeted to robot morphologies. Although such data provides rich semantic and kinematic priors, it fails to capture the nuances of whole-body tracking controllers, including balance, contact dynamics, actuation limits, and controller-specific failure modes. A

  Source
• CoorDex: Coordinating Body and Hand Priors for Continuous Dexterous Humanoid Loco-Manipulation

  2026-06-22 - Research Publication - CoorDex coordinates body and hand priors for simulated continuous humanoid loco-manipulation. Quantitative task evaluation uses G1 with WUJI hands in Isaac Lab; separate physical G1 with Dex3-1 visualizations replay recorded joint trajectories.

  Source
• LP-NavOA: Integrated Local Navigation and Obstacle Avoidance for Humanoid Robots under Limited Perception

  2026-06-22 - Research Publication - LP-NavOA combines a range-conditioned locomotion backbone with a learned local steering interface for obstacle bypass and sparse-goal recovery. It evaluates G1 navigation in MuJoCo and separately demonstrates bounded physical wall-bypass trials without continuous joystick steering.

  Source
• Learning Dexterous Manipulation Using Contact Wrench Guidance From Human Demonstration

  2026-06-22 - Research Publication - CHORD learns contact-rich dexterous manipulation from human demonstrations by matching object-centric contact wrenches. It evaluates a large simulated task library and transfers selected policies to a fixed-base Dexmate robot with two Sharpa hands.

  Source
• VQActFlow: Vector-Quantized Action Mode Steering for Multi-Task Robot Manipulation

  2026-06-19 - Research Publication - Multi-task robot manipulation policies are challenging to learn from demonstration because traditionally a single network must select among qualitatively different action modes from a multimodal demonstration distribution, conditioned on language and visual context.

  Source
• Duet: Dual-Robot Understanding via Efficient Teaching

  2026-06-18 - Research Publication - DUET learns dual-robot collaboration through a shared action representation and parallel control heads. Four contact-rich physical tasks pair a Unitree G1 with a Dexmate Vega1, using VR teaching and embodiment-specific low-level control.

  Source
• HALOMI: Learning Humanoid Loco-Manipulation with Active Perception from Human Demonstrations

  2026-06-17 - Research Publication - Human demonstrations, which can be collected at scale and naturally capture active hand-eye coordination, are a promising data source for learning humanoid loco-manipulation.

  Source
• MagicSim: A Unified Infrastructure for Executable Embodied Interaction

  2026-06-16 - Research Publication - MagicSim proposes a shared runtime for simulated robot control, skills and planning, combining executable robot embodiments with multiple drivers. Its registry includes Unitree G1 variants and a Vega-1P mobile manipulator with Sharpa Wave hands.

  Source
• Steering Generative Reinforcement Learning into Stable Robotic Controller

  2026-06-15 - Research Publication - Diffusion and flow-based generative policies provide a powerful policy class for reinforcement learning by inducing rich stochastic exploration through iterative action generation.

  Source
• ADAPT: Analytical Disturbance-Aware Policy Training for Humanoid Locomotion

  2026-06-15 - Research Publication - Humanoids deployed in human-centered environments must handle force-interactive tasks, where external contacts introduce unexpected disturbances that disrupt locomotion accuracy and stability.

  Source
• Energy-Efficient Arm Reaching for a Humanoid Robot via Deep Reinforcement Learning with Identified Power Models

  2026-06-14 - Research Publication - Humanoid robots performing in-field manipulation tasks, such as robotic apple harvesting, face severe energy constraints that directly limit the number of reaching motions that can be executed per battery charge. This paper presents an end-to-end, energy-aware reinforcement learning framework for the 7-degree-of-freedom left arm of the Unitree~G1 humanoid robot, combining a physics-based, experimentally identified electrical power model with a Soft Actor-Critic (SAC) policy trained in a Pinocchio-based rigid-body d

  Source
• Identification of a Physics-Based Electrical Power Consumption Model for the Unitree G1 Humanoid Arm

  2026-06-14 - Research Publication - Accurate prediction of electrical power consumption is essential for energy-aware motion planning, battery management, and thermal monitoring in battery-powered humanoid robots.

  Source
• Whole-Body Impedance Model Predictive Control for Safe Physical Human--Robot Interaction on Floating-Base Platforms

  2026-06-12 - Research Publication - Floating-base robots must balance under rigid contact constraints while interacting safely with humans.

  Source
• FAST_LIO_LOCALIZATION_HUMANOID: Localization by LiDAR for Humanoid

  2026-06-12 - Research Publication - Open-sources a LiDAR localization stack adapted for humanoid platforms such as Unitree G1-style robots. Builds on FAST-LIO with humanoid-oriented deployment changes and ROS/C++ prerequisites. Adds practical perception/navigation infrastructure for stable humanoid operation in unstructured environments. Open-sources a LiDAR localization stack adapted for humanoid platforms such as Unitree G1-style robots. Builds on FAST-LIO with humanoid-oriented deployment changes and ROS/C++ prerequisites. Adds practical perception/navigation infrastructure for stable humanoid operation in unstructured environments. Concrete GitHub release for humanoid perception and navigation support infrastructure.

  Source
• Critic Architecture Matters: Dual vs. Unified Critics for Humanoid Loco-Manipulation

  2026-06-10 - Research Publication - Multi-objective reinforcement learning for humanoid robots must coordinate locomotion and manipulation within a single policy.

  Source
• RoboNaldo: Accurate, Stable and Powerful Humanoid Soccer Shooting via Motion-Guided Curriculum Reinforcement Learning

  2026-06-09 - Research Publication - Elite humanoid soccer shooting requires whole-body stability, high-impulse whole-body interactions, and accuracy to targets.

  Source
• MARCH: Model-Assisted Reinforcement Learning for the Perceptive Control of Humanoids over Sparse Footholds

  2026-06-09 - Research Publication - Perceptive bipedal locomotion over sparse terrain remains a difficult challenge: model-based methods are precise but brittle to uncertainty, while model-free methods are robust but struggle to discover the precise, constrained motions required for safety-critical locomotion where small errors can cause catastrophic failures.

  Source
• MotionWAM: Towards Foundation World Action Models for Real-Time Humanoid Loco-Manipulation

  2026-06-08 - Research Publication - World Action Models (WAMs) couple a video dynamics prior to the policy and have shown encouraging results on tabletop manipulation, but iterative denoising over high-dimensional video-action latents leaves them too slow for real-time humanoid loco-manipulation.

  Source
• UniReLo: Learning a Unified Humanoid Policy from Fall Recovery to Locomotion across Diverse Terrains

  2026-06-08 - Research Publication - Humanoid robots can fall on slopes, gravel, and uneven ground in unstructured environments.

  Source
• Cooperative Long Rope Skipping via Multi-Agent Reinforcement Learning

  2026-06-06 - Research Publication - Humans exhibit remarkable motor agility, enabling a wide range of dynamic skills such as running and jumping, which highlights the great potential of humanoid robots for athletic locomotion.

  Source
• Anchor Lab Dataset: Real-World Measurements for Sim-to-Real Calibration

  2026-06-05 - Research Publication - Releases a Hugging Face robotics dataset for grounding simulation calibration in real physical measurements.

  Source
• Predictive Style Matching: Natural and Robust Humanoid Locomotion

  2026-06-05 - Research Publication - Reinforcement learning has become the prevailing approach to humanoid locomotion control: policies transfer reliably from simulation to hardware and recover gracefully from disturbances.

  Source
• HANDOFF: Humanoid Agentic Task-Space Whole-Body Control via Distilled Complementary Teachers

  2026-06-04 - Research Publication - For a humanoid robot to be deployed in the real world, the choice of command space (i.

  Source
• LEGS: Fine-Tuning Teleop-Free VLAs for Humanoid Loco-manipulation in an Embodied Gaussian Splatting World

  2026-06-03 - Research Publication - Introduces a teleoperation-free VLA fine-tuning pipeline for humanoid loco-manipulation in an embodied Gaussian Splatting world. Uses MuJoCo physics, photorealistic 3DGS backgrounds, SAM3D objects, and motion primitives to generate labeled demos. Reports Unitree G1 pick-and-place results with re-rendered data enabling cheaper robustness training. Introduces a teleoperation-free VLA fine-tuning pipeline for humanoid loco-manipulation in an embodied Gaussian Splatting world. Uses MuJoCo physics, photorealistic 3DGS backgrounds, SAM3D objects, and motion primitives to generate labeled demos. Reports Unitree G1 pick-and-place results with re-rendered data enabling cheaper robustness training. Concrete humanoid VLA project page with arXiv paper and sim-to-real loco-manipulation evaluation.

  Source
• GRAIL: Generating Humanoid Loco-Manipulation from 3D Assets and Video Priors

  2026-06-03 - Research Publication - Scaling humanoid loco-manipulation requires robot-compatible demonstrations across diverse objects, whole-body motions, and scene geometries, but teleoperation and motion capture are difficult to scale because each collection depends on physical setups, instrumented actors, and robot operation.

  Source
• Flash-WAM: Modality-Aware Distillation for World Action Models

  2026-06-03 - Research Publication - World-action models (WAMs) jointly generate future video and robot actions through iterative diffusion, achieving strong performance on manipulation benchmarks but requiring tens of denoising steps, a cost that precludes real-time control.

  Source
• M3imic: Learning a Versatile Whole-Body Controller for Multimodal Motion Mimicking

  2026-06-03 - Research Publication - Building a general-purpose whole-body controller is essential for enabling diverse motion capabilities in humanoid robots across a wide range of downstream tasks, including locomotion and loco-manipulation.

  Source
• CoRe-MoE: Contrastive Reweighted Mixture of Experts for Multi-Terrain Humanoid Locomotion with Gait Adaptation

  2026-06-03 - Research Publication - Humans primarily rely on walking and running to traverse complex terrains.

  Source
• Humanoid-GPT: Scaling Data and Structure for Zero-Shot Motion Tracking

  2026-06-02 - Research Publication - Releases a GPT-style humanoid motion Transformer trained on a 2B-frame retargeted motion corpus for zero-shot motion tracking. Repository includes inference, evaluation, MuJoCo/MJX simulation, released checkpoints, and Unitree G1 deployment paths. Project page reports real-time whole-body control and zero-shot tracking on Unitree G1 across everyday home and high-dynamic motion scenarios. Releases a GPT-style humanoid motion Transformer trained on a 2B-frame retargeted motion corpus for zero-shot motion tracking. Repository includes inference, evaluation, MuJoCo/MJX simulation, released checkpoints, and Unitree G1 deployment paths. Project page reports real-time whole-body control and zero-shot tracking on Unitree G1 across everyday home and high-dynamic motion scenarios. High-signal humanoid code release with billion-scale motion data, released checkpoints, and real Unitree G1 deployment support.

  Source
• Bionic Human-Motion Style Transfer for Physically Executable Whole-Body Control of Humanoid Robots

  2026-06-02 - Research Publication - Expressive whole-body motion is important for humanoid robots operating in human environments, where robots are expected to move stably while presenting readable and adjustable body behaviors.

  Source
• Human2Humanoid: Physics-Aware Cross-Morphology Motion Retargeting for Humanoid Robots

  2026-06-02 - Research Publication - Retargeting human motion to humanoid robots is critical for teleoperation, imitation learning and human-robot interaction. However, it remains challenging because of substantial morphological discrepancies between humans and robots, including differences in skeletal topology, limb proportions and degrees of freedom, as well as the scarcity of paired motion data. This paper presents Human2Humanoid, an unsupervised mot

  Source
• Global-Local Attention Decomposition for Terrain Encoding in Humanoid Perceptive Locomotion

  2026-05-30 - Research Publication - Although reinforcement learning has significantly advanced humanoid locomotion, perceptive policies still struggle on sparse-foothold terrain and constrained environments. Success in these scenarios requires both broad terrain awareness and precise foothold selection, two perceptual roles that conventional encoders often entangle. To address this challenge, we propose Global-Local Attention Decomposition (GLAD) for t

  Source
• Gaze2Act: Gaze-Conditioned Vision-Language-Action Policies for Interactive Robot Manipulation

  2026-05-28 - Research Publication - Vision-Language-Action (VLA) models have recently shown strong potential for robot learning by following language instructions. However, in practice, language alone is often insufficient to precisely convey human intent. It is difficult to describe which exact object to interact with among similar candidates, where to act on the object, or how the target may change during execution. To address this limitation, we pro

  Source
• Learning to Evolve: Multi-modal Interactive Fields for Robust Humanoid Navigation in Dynamic Environments

  2026-05-21 - Research Publication - Safe manipulation-oriented navigation for humanoid robots requires scene memory that remains reliable under locomotion-induced perceptual distortion, environmental changes, and interaction-level geometric safety constraints. Existing semantic mapping and scene-graph systems are difficult to deploy directly in this setting because they often assume stable camera trajectories, static environments, or coarse object geom

  Source
• State-dependent AMP paper unifies G1 walking, running, and fall recovery

  2026-05-18 - Research Publication - This paper covers State-dependent AMP paper unifies G1 walking, running, and fall recovery for G1 in Reinforcement learning.

  Source
• PRIME: Physically-consistent Robotic Inertial and Motion Estimation for Legged and Humanoid Robots

  2026-05-17 - Research Publication - Humanoid and legged robots interact with the environment through intermittent contacts, making accurate motion estimation fundamentally dependent on reasoning about contact dynamics. However, standard sensing pipelines-whether based on onboard proprioception with Extended Kalman Filters (EKFs) or external motion capture systems-recover only kinematics, while contact forces, contact timing, and inertial parameters rem

  Source
• Tuesday-Labs/maya

  2026-05-13 - Research Publication - The voice and brain for Maya, an AI-powered Unitree G1 robot.

  Source
• IMU teleoperation paper validates real-time whole-body control on Unitree G1

  2026-05-12 - Research Publication - Stable, low-latency whole-body teleoperation of humanoid robots is an open research challenge, complicated by kinematic mismatches between human and robot morphologies, accumulated inertial sensor noise, non-trivial control latency, and persistent sim-to-real transfer gaps.

  Source
• Explicit Stair Geometry Conditioning for Robust Humanoid Locomotion

  2026-05-11 - Research Publication - Robust humanoid stair climbing remains challenging due to geometric discontinuities, sensitivity to step height variations, and perception uncertainty in real-world environments. Existing learning-based locomotion policies often rely on implicit terrain representations or blind proprioceptive feedback, limiting their ability to generalize across varying stair geometries and to anticipate required gait adjustments. Th

  Source
• Constraint-Enhanced Reinforcement Learning Based on Dynamic Decoupled Spherical Radial Squashing

  2026-05-05 - Research Publication - When deploying reinforcement learning policies to physical robots, actuator rate constraints -- hard limits on how fast each joint can move per control step -- are unavoidable.

  Source
• Action Agent: Agentic Video Generation Meets Flow-Constrained Diffusion

  2026-05-02 - Research Publication - We present Action Agent, a two-stage framework that unifies agentic navigation video generation with flow-constrained diffusion control for multi-embodiment robot navigation. In Stage I, a large language model (LLM) acts as an orchestration module that selects video diffusion models, refines prompts through iterative validation, and accumulates cross-task memory to synthesize physically plausible first-person navigat

  Source
• Foundation-model grasping pipeline streams commands to Unitree G1

  2026-04-19 - Research Publication - Deploying a humanoid robot to manipulate a new object has traditionally required one to two days of effort: data collection, manual annotation, 3D model acquisition, and model training.

  Source
• Whole-body locomotion paper deploys terrain-aware motion generation on Unitree G1

  2026-04-19 - Research Publication - Whole-body humanoid locomotion is challenging due to high-dimensional control, morphological instability, and the need for real-time adaptation to various terrains using onboard perception.

  Source
• Renforce-Dynamics/TableTennis_Deploy

  2026-04-17 - Research Publication - This repository is for deploying table tennis skills on the Unitree G1 robot. It is derived from the "RoboMimic_Deploy" repository and includes related skills from the original project.

  Source
• CLAW: Composable Language-Annotated Whole-body Motion Generation

  2026-04-13 - Research Publication - Training language-conditioned whole-body controllers for humanoid robots demands large-scale motion-language datasets.

  Source
• FlashSAC: Fast and Stable Off-Policy Reinforcement Learning for High-Dimensional Robot Control

  2026-04-06 - Research Publication - FlashSAC scales off-policy reinforcement learning with larger networks, high-throughput data and bounded optimization norms. It benchmarks diverse simulation tasks and transfers blind locomotion policies to a physical 29-DoF Unitree G1, including flat motion and stair climbing.

  Source
• LefferTrochez/Sim2Real-Control-Framework-for-Unitree-G1-Simulink-ROS2-MuJoCo

  2026-04-03 - Research Publication - A MATLAB/Simulink-based Sim2Real framework for the Unitree G1 humanoid robot, integrating MuJoCo and ROS 2 in a unified workflow for controller development, simulation, visualization, and real-robot deployment.

  Source
• Learning Humanoid Navigation from Human Data

  2026-04-01 - Research Publication - We present EgoNav, a system that enables a humanoid robot to traverse diverse, unseen environments by learning entirely from 5 hours of human walking data, with no robot data or finetuning.

  Source
• NVIDIA releases synthetic G1 Locomanipulation Dataset v1

  2026-04-01 - Research Publication - This dataset covers NVIDIA releases synthetic G1 Locomanipulation Dataset v1 for G1 in dexterity.

  Source
• Jiarui-Xie/AMP_Running_baseline

  2026-04-01 - Research Publication - Locomotion training codebase for Unitree G1 competing in the 2026 Beijing Yizhuang Half Marathon Robot Race. Built on Isaac Lab with Adversarial Motion Priors (AMP), motion imitation, and hardware-ready deployment.

  Source
• SMASH: Humanoid Table Tennis with Egocentric Vision

  2026-04-01 - Research Publication - SMASH demonstrates consecutive table-tennis strikes using onboard vision and learned whole-body control.

  Source
• Unitree expands the UnifoLM_G1_Dex1_Dataset collection for G1 dexterous manipulation

  2026-03-31 - Research Publication - Unitree published an expanded UnifoLM_G1_Dex1_Dataset collection on Hugging Face for the G1 platform, grouping multiple dexterous manipulation datasets under a single first-party release.

  Source
• Active stereo-camera study benchmarks sensor mixes on Unitree G1

  2026-03-30 - Research Publication - The complexity of teaching humanoid robots new tasks is one of the major reasons hindering their widespread adoption in the industry.

  Source
• Chasing Autonomy: Dynamic Retargeting and Control Guided RL for Performant and Controllable Humanoid Running

  2026-03-26 - Research Publication - The paper combines dynamic retargeting with control-guided reinforcement learning for performant, controllable humanoid running. Hardware experiments deploy the policy on Unitree G1 and report running speeds up to 3.

  Source
• Make Tracking Easy: Neural Motion Retargeting for Humanoid Whole-body Control

  2026-03-23 - Research Publication - Humanoid robots require diverse motor skills to integrate into complex environments, but bridging the kinematic and dynamic embodiment gap from human data remains a major bottleneck.

  Source
• AGILE: A Comprehensive Workflow for Humanoid Loco-Manipulation Learning

  2026-03-20 - Research Publication - Recent advances in reinforcement learning (RL) have enabled impressive humanoid behaviors in simulation, yet transferring these results to new robots remains challenging.

  Source
• alexzh3/OpenHumanoid

  2026-03-20 - Research Publication - OpenHumanoid provides voice-controlled locomotion for a Unitree G1 humanoid robot. Won 2nd place at ETH Robotics Club hackathon.

  Source
• maedmatt/DreamMotion

  2026-03-19 - Research Publication - LLM agent that generates Unitree G1 humanoid robot motions from natural language.

  Source
• ECHO paper runs language-to-motion whole-body control on Unitree G1

  2026-03-17 - Research Publication - We present ECHO, an edge--cloud framework for language-driven whole-body control of humanoid robots.

  Source
• nv-tlabs/kimodo

  2026-03-16 - Research Publication - Official implementation of Kimodo, a kinematic motion diffusion model for high-quality human(oid) motion generation.

  Source
• sesteban951/deploy_robot

  2026-03-16 - Research Publication - Deploy code for the Unitree G1 robot.

  Source
• LATENT deploys learned humanoid tennis skills on Unitree G1

  2026-03-13 - Research Publication - Human athletes demonstrate versatile and highly-dynamic tennis skills to successfully conduct competitive rallies with a high-speed tennis ball.

  Source
• InterReal

  2026-03-12 - Research Publication - Presents a paper on a framework for learning human-object interaction (HOI) skills in robots. Facilitates tracking of reference motions for humanoid robots, improving fine-grained interactive skill learning. Enables applications in HOI tasks such as box-picking and box-pushing, demonstrating effectiveness on the Unitree G1 robot. InterReal: A Unified Physics-Based Imitation Framework for Learning Human-Object Interaction Skills Dayang Liang, Yuhang Lin, Xinzhe Liu, Jiyuan Shi, Yunlong Liu, Chenjia Bai https://t.co/QpNPWMYKWU Explores practical applications of learning human-object interaction for humanoid robots.

  Source
• Psi0: An Open Foundation Model Towards Universal Humanoid Loco-Manipulation

  2026-03-12 - Research Publication - Psi0 separates human-video representation learning from humanoid action learning and evaluates eight long-horizon loco-manipulation tasks on physical Unitree G1.

  Source
• ZeroWBC learns natural whole-body control for Unitree G1 directly from egocentric human video

  2026-03-10 - Research Publication - Achieving versatile and natural whole-body humanoid interaction control remains challenging due to the high cost of whole-body teleoperation data.

  Source
• Cross-Hand Latent Representation for Vision-Language-Action Models

  2026-03-10 - Research Publication - XL-VLA evaluates a physical PaXini DexH13 in real-world bimanual xArm experiments spanning ten manipulation tasks. Across ten trials per task, the reported DexH13 mean success rate is 0.78 versus 0.35 for the direct-action baseline; the broader training corpus contains two million state-action pairs across four hand embodiments.

  Source
• Unitree releases the UnifoLM_WBT_Dataset for humanoid whole-body teleoperation

  2026-03-05 - Research Publication - Unitree publicly released the UnifoLM_WBT_Dataset as a real-world humanoid whole-body teleoperation dataset through its Hugging Face collection.

  Source
• Rhythm deploys dual-humanoid interactive whole-body control on Unitree G1 robots

  2026-03-03 - Research Publication - Realizing interactive whole-body control for multi-humanoid systems is critical for unlocking complex collaborative capabilities in shared environments.

  Source
• ULTRA paper demonstrates autonomous whole-body loco-manipulation on Unitree G1

  2026-03-03 - Research Publication - Achieving autonomous and versatile whole-body loco-manipulation remains a central barrier to making humanoids practically useful.

  Source
• adrialemany/g1man

  2026-03-02 - Research Publication - Controlling the Unitree G1 robot in a simple way, using a UI, or scripts. Manipulation, pointcloud and simulation

  Source
• heyjiacheng/teleoperate_G1

  2026-03-02 - Research Publication - A guide to teleoperate unitree G1 robot in RPL Lab.

  Source
• Humanoid Robots as First Assistants in Endoscopic Surgery

  2026-02-27 - Research Publication - Humanoid robots have become a focal point of technological ambition, with claims of surgical capability within years in mainstream discourse.

  Source
• physical-superintelligence-lab/Psi0

  2026-02-27 - Research Publication - [RSS26'] Welcome to Psi-Zero, a Humanoid VLA towards Universal Humanoid Intelligence.

  Source
• ozkannceylan/humanoid_vla

  2026-02-26 - Research Publication - End-to-end VLA humanoid robot in MuJoCo: ACT model on Unitree G1 with ROS 2 NL interface and OOD generalization study.

  Source
• UAI-BIOARABA/saaki-ros2-yolo

  2026-02-25 - Research Publication - Unitree G1 - Object detection with YOLO and ROS2 for the real robot.

  Source
• experientialtech/g1-moves

  2026-02-23 - Research Publication - Motion capture clips for Unitree G1 humanoid robot

  Source
• z4ziggy/z4rtc

  2026-02-19 - Research Publication - A full-featured command-line controller for the Unitree G1/Go2 robots, using WebRTC.

  Source
• Perceptive Humanoid Parkour: Chaining Dynamic Human Skills via Motion Matching

  2026-02-17 - Research Publication - While recent advances in humanoid locomotion have achieved stable walking on varied terrains, capturing the agility and adaptivity of highly dynamic human motions remains an open challenge.

  Source
• uleroboticsgroup/rl_hnav

  2026-02-14 - Research Publication - Modular ROS 2 workspace for hybrid navigation of Unitree G1: RL locomotion (MuJoCo), SLAM, Nav2, TF bridging, and real robot deployment.

  Source
• shivangvijay/Deep-Skill-Graph-Based-Humanoid-Navigation-DSG-HN

  2026-02-13 - Research Publication - We propose extending DSG-style skill-graph planning to a high-DoF humanoid robot (Unitree G1) in structured indoor environments. Our objective is to evaluate whether graph-based abstraction scales to dynamically constrained humanoid locomotion and to introduce stability-aware and uncertainty-aware extensions suitable for whole-body systems.

  Source
• APEX learns adaptive high-platform traversal on a 29-DoF Unitree G1

  2026-02-11 - Research Publication - Humanoid locomotion has advanced rapidly with deep reinforcement learning (DRL), enabling robust feet-based traversal over uneven terrain.

  Source
• Embodiment-Aware Generalist Specialist Distillation for Unified Humanoid Whole-Body Control

  2026-02-03 - Research Publication - Humanoid Whole-Body Controllers trained with reinforcement learning (RL) have recently achieved remarkable performance, yet many target a single robot embodiment.

  Source
• HUSKY: Humanoid Skateboarding System via Physics-Aware Whole-Body Control

  2026-02-03 - Research Publication - While current humanoid whole-body control frameworks predominantly rely on the static environment assumptions, addressing tasks characterized by high dynamism and complex interactions presents a formidable challenge.

  Source
• NPCLEI/KungFuAthleteBot

  2026-02-02 - Research Publication - A highly dynamic martial arts motion dataset (848 samples, Ground/Jump subsets) for humanoid robots, with GMR retargeting, height adjustment, and RL training pipeline (PPO & FastSAC) for Unitree G1, featuring fall recovery and sim-to-real deployment.

  Source
• Aseke314/unitree-g1-navigation

  2026-02-01 - Research Publication - Autonomous navigation for Unitree G1 humanoid robot in MuJoCo using RL policy and virtual lidar

  Source
• Uvesh-patel/unitree-g1-docker-sdk

  2026-01-23 - Research Publication - Ready-to-use Docker environment for Unitree G1 humanoid robot development. Includes Python SDK, Cyclone DDS, and examples. Works on Windows and macOS.

  Source
• HumanoidVLM uses Unitree G1 for vision-language-guided impedance control research

  2026-01-21 - Research Publication - Humanoid robots must adapt their contact behavior to diverse objects and tasks, yet most controllers rely on fixed, hand-tuned impedance gains and gripper settings.

  Source
• FRoM-W1: Towards General Humanoid Whole-Body Control with Language Instructions

  2026-01-19 - Research Publication - Humanoid robots are capable of performing various actions such as greeting, dancing and even backflipping.

  Source
• TWIST2

  2026-01-03 - Research Publication - Releases a repository for a humanoid data collection system designed for teleoperation with Unitree G1. Facilitates real and simulated control through various scripts and integration with a VR headset. Supports sim-to-real transfer and manipulation tasks in humanoid robots. TWIST2's focus on humanoid data collection aligns with robotics research in teleoperation and sim-to-real applications.

  Source
• NVIDIA GR00T N1.6 humanoid foundation model

  2025-12-15 - Research Publication - NVIDIA Research published GR00T N1.6 as an improved foundation model for generalist humanoid robots. The page describes long-horizon reasoning, dexterity, and multitasking experiments on bimanual YAM, Agibot Genie-1, and Unitree G1 robots.

  Source
• From Generated Human Videos to Physically Plausible Robot Trajectories

  2025-12-04 - Research Publication - Ships a pipeline that maps generated RGB video into 4D human reconstruction, retargeted humanoid keypoints, and IsaacGym policy execution on G1. Belongs to human-video to robot trajectory transfer. Matters because it expands demonstration supply beyond teleop by using generated human motion as a source for executable robot trajectories. Ships a pipeline that maps generated RGB video into 4D human reconstruction, retargeted humanoid keypoints, and IsaacGym policy execution on G1. Belongs to human-video to robot trajectory transfer. Matters because it expands demonstration supply beyond teleop by using generated human motion as a source for executable robot trajectories. Project page: https://genmimic.github.io/. Confidence: direct artifact link in main post.

  Source
• kairos-agi/kairos

  2025-12-02 - Research Publication - Official code for world model Kairos

  Source
• Modality-Augmented Fine-Tuning of Foundation Robot Policies for Cross-Embodiment Manipulation on GR1 and G1

  2025-12-01 - Research Publication - This simulation study augments GR00T policies with contact and depth signals across Fourier GR1 and Unitree G1 embodiments. It uses RoboSuite/RoboCasa rollouts and releases simulated G1 apple-to-bowl demonstrations with proprioception and physics-engine contact measurements.

  Source
• METIS: Multi-Source Egocentric Training for Integrated Dexterous Vision-Language-Action Model

  2025-11-21 - Research Publication - METIS combines multi-source egocentric action data with motion-aware dynamics and reasoning for dexterous manipulation. Physical tests use Unitree G1 with Inspire hands.

  Source
• amazon-far/holosoma

  2025-11-13 - Research Publication - with algorithms like PPO and FastSAC. ## Features - **Multi-simulator support**: IsaacGym, IsaacSim, MuJoCo Warp (MJWarp), and MuJoCo (inference only) - **Multiple RL algorithms**: PPO and FastSAC - **Robot support**: Unitree G1 and Booster T1 humanoids - **Task types**: Locomotion (velocity tracking) and whole-body tracking - **Sim-to-sim and sim-to-real deployment**: Shared inference pipeline across simulation and real robot control - **Motion retargeting**: Convert human motion capture data to robot

  Source
• AIM-Intelligence/AIM-Robotics

  2025-10-27 - Research Publication - Unitree G1 Robot

  Source
• Layered-control architecture paper demonstrates robust stairs and ledges on Unitree G1

  2025-10-16 - Research Publication - This paper covers Layered-control architecture paper demonstrates robust stairs and ledges on Unitree G1 for G1 in Reinforcement learning.

  Source
• DemoHLM shows generalizable loco-manipulation on Unitree G1 from a single simulated demonstration

  2025-10-13 - Research Publication - Loco-manipulation is a fundamental challenge for humanoid robots to achieve versatile interactions in human environments.

  Source
• Humanoid Everyday releases open-world manipulation dataset collected on Unitree G1

  2025-10-09 - Research Publication - This dataset covers Humanoid Everyday releases open-world manipulation dataset collected on Unitree G1 for G1 in dexterity.

  Source
• RobotDancing paper validates long-horizon humanoid motion tracking on Unitree H1

  2025-09-25 - Research Publication - Long-horizon, high-dynamic motion tracking on humanoids remains brittle because absolute joint commands cannot compensate model-plant mismatch, leading to error accumulation.

  Source
• RuN paper demonstrates natural walk-run transitions on Unitree G1

  2025-09-25 - Research Publication - This paper covers RuN paper demonstrates natural walk-run transitions on Unitree G1 for G1 in Robot learning.

  Source
• RoMoCo: Robotic Motion Control Toolbox for Reduced-Order Model-Based Locomotion on Bipedal and Humanoid Robots

  2025-09-23 - Research Publication - We present RoMoCo, an open-source C++ toolbox for the synthesis and evaluation of reduced-order model-based planners and whole-body controllers for bipedal and humanoid robots.

  Source
• knoxsbyte/argos

  2025-09-23 - Research Publication - A fully end-to-end framework for coordinating swarms of Unitree G1 humanoid robots to autonomously clean rooms

  Source
• HDMI: Learning Interactive Humanoid Whole-Body Control from Human Videos

  2025-09-20 - Research Publication - HDMI extracts human and object motion from videos, trains joint robot-object tracking policies in simulation, and transfers them to physical Unitree G1. The experiments include repeated door traversal, box handling and other whole-body interaction tasks.

  Source
• shanpenghui/RoboMimicDeploy_G1

  2025-08-14 - Research Publication - RoboMimicDeploy_G1 is a policy deployment and control framework for the Unitree G1 robot, supporting both Mujoco simulation and real-time control on the physical robot.

  Source
• hangxu811/Unitree_G1_pose_control-through-UI

  2025-08-14 - Research Publication - This is a motion-design assistant tool for the Unitree G1 humanoid robot. It enables users to intuitively design and adjust poses in a 3D environment, then send them to the control panel so the real G1 can accurately reproduce the same posture.

  Source
• Ericcsr/G1_localization

  2025-08-05 - Research Publication - Localization toolbox for Unitree G1 humanoid robot

  Source
• YanjieZe/GMR

  2025-08-03 - Research Publication - [ICRA 2026] GMR: General Motion Retargeting. Retarget human motions into diverse humanoid robots in real time on CPU. Retargeter for TWIST.

  Source
• GalacTechNyc/unitree-g1-autonomous

  2025-07-28 - Research Publication - 🤖 Fully autonomous navigation system for Unitree G1 humanoid robot using AI-powered visual analysis with Google Gemini API

  Source
• JeffrinSam/JeffrinSam_G1

  2025-07-22 - Research Publication - Synthetic Data Generation -Generate synthetic joint data for Unitree H1/G1 robots via imitation learning in Isaac Lab. Simulate and visualize in Isaac Sim, enhance with Cosmos for realism, and combine with teleop data to fine-tune Gr00t N1.5 for pick-and-place tasks, tested in Isaac Sim(Inference)

  Source
• wkooks/LIOrf_ros2_G1

  2025-07-22 - Research Publication - LIOrf ROS2 package for Unitree G1 robot

  Source
• BrainCoTech/unitree-g1-brainco-hand

  2025-07-16 - Research Publication - Tutorial on adapting BrainCo Revo2 hands for Unitree G1 robot.

  Source
• UniTracker universal whole-body motion tracker is evaluated on Unitree G1

  2025-07-10 - Research Publication - Achieving expressive and generalizable whole-body motion control is essential for deploying humanoid robots in real-world environments.

  Source
• yifeichen2024/G1_deploy

  2025-07-02 - Research Publication - Deploy the policy on the unitree G1 robot

  Source
• NVIDIA GR00T N1.5 humanoid foundation model

  2025-06-11 - Research Publication - NVIDIA Research published GR00T N1.5 as an upgraded open foundation model for generalist humanoid robots. The page reports improved simulated benchmark performance, stronger language following on Fourier GR-1 experiments, and post-training on Unitree G1 teleoperation episodes.

  Source
• NVIDIA releases G1 fruit pick-and-place teleoperation dataset for GR00T workflows

  2025-06-10 - Research Publication - This dataset covers NVIDIA releases G1 fruit pick-and-place teleoperation dataset for GR00T workflows for G1 in Human-robot interaction.

  Source
• mujocolab/mjlab

  2025-06-10 - Research Publication - Isaac Lab API, powered by MuJoCo-Warp, for RL and robotics research

  Source
• Galbot open-sources OpenWBT whole-body teleoperation for humanoid robots

  2025-06-09 - Research Publication - This repo covers Galbot open-sources OpenWBT whole-body teleoperation for humanoid robots for Galbot G1 in Human-robot interaction.

  Source
• Jalil32/g1-isaac-groot-n1

  2025-05-28 - Research Publication - This project aims to fine-tune and deploy NVIDIA's Isaac GR00T N1 model to the unitree G1 humanoid robot

  Source
• Gait-conditioned reinforcement-learning paper validates multi-gait locomotion on Unitree G1

  2025-05-27 - Research Publication - We present a unified gait-conditioned reinforcement learning framework that enables humanoid robots to perform standing, walking, running, and smooth transitions within a single recurrent policy.

  Source
• Learning a Unified Policy for Position and Force Control in Legged Loco-Manipulation

  2025-05-27 - Research Publication - Adds a cross-platform Unitree manipulation/control artifact spanning B2 and G1.

  Source
• jiachengliu3/OpenWBC

  2025-05-22 - Research Publication - VR-based Robot Teleoperation and Data Collection System for Humanoid Whole-Body VLA (Unitree G1)

  Source
• huangshk/Unitree-G1-Control-Replay

  2025-05-21 - Research Publication - Unitree G1 Robot Control & Replay Platform

  Source
• TeleOpBench: A Simulator-Centric Benchmark for Dual-Arm Dexterous Teleoperation

  2025-05-19 - Research Publication - TeleOpBench compares motion-capture, vision, VR and exoskeleton interfaces in30 simulated bimanual tasks, then compares trends with physical dual-arm experiments.

  Source
• yueruyin/robot-unitree-g1

  2025-05-12 - Research Publication - 基于mujoco仿真环境对unitree g1机器操作的研究和学习

  Source
• yediong/IK-humanoid

  2025-04-19 - Research Publication - Using inverse kinematics (IK) solving to achieve partial hand movements for the Unitree G1 robot, based on unitree-rl-gym and IsaacGym.

  Source
• Humanoid Agent via Embodied Chain-of-Action Reasoning with Multimodal Foundation Models for Zero-Shot Loco-Manipulation

  2025-04-13 - Research Publication - Humanoid-COA decomposes natural-language instructions into grounded action sequences and evaluates physical H1-2 and G1 loco-manipulation in an open area and apartment. The framework combines visual scene interpretation, affordance reasoning, manipulation primitives and navigation.

  Source
• aliasrobotics/cai

  2025-03-31 - Research Publication - Cybersecurity AI (CAI), the framework for AI Security

  Source
• ashishmiryalkar/unitree_walking

  2025-03-13 - Research Publication - unitree g1 robot walking using omniverse, Isaaclab, PPO Reinforcement learning.

  Source
• ioloizou/g1_locomotion

  2025-02-27 - Research Publication - Humanoid Robot Locomotion Stack for Unitree G1 Using Convex Model Predictive Control

  Source
• HOMIE: Humanoid Loco-Manipulation with Isomorphic Exoskeleton Cockpit

  2025-02-25 - Research Publication - Releases a repository for HUMIE: Humanoid Loco-Manipulation with Isomorphic Exoskeleton Cockpit. Covers frameworks for locomotion and manipulation tasks in humanoid robots. Targets humanoid robot training and operations using reinforcement learning. Spent half an hour to modify the training code of HOMIE (https://t.co/ZjcSUifB47) to train another kind of full-size humanoid, Unitree H1-2, which is much heavier than Unitree G1. Without some modific The artifact focuses on training humanoid robots for loco-manipulation tasks.

  Source
• aukilabs/auki_robotics_g1_humanoid_ros2

  2025-02-13 - Research Publication - ROS2 packages for Unitree G1 Humanoid

  Source
• HoST paper deploys diverse-posture standing-up control on Unitree G1

  2025-02-12 - Research Publication - Standing-up control is crucial for humanoid robots, with the potential for integration into current locomotion and loco-manipulation systems, such as fall recovery.

  Source
• ASAP paper aligns simulation and real-world physics for agile Unitree G1 skills

  2025-02-03 - Research Publication - Humanoid robots hold the potential for unparalleled versatility in performing human-like, whole-body skills.

  Source
• Ahmad0Aldaher/Aligator_Unitree_G1

  2025-01-26 - Research Publication - Modeling the walking behavior of the Unitree G1 robot using the Aligator trajectory optimization library.

  Source
• Ahmad0Aldaher/Unitree_G1_Stabilization

  2025-01-26 - Research Publication - This Git repository provides a Python script using the Pinocchio library and the Unitree G1 humanoid robot model

  Source
• kyavuzkurt/g1_description

  2024-12-03 - Research Publication - ROS2 Package for Unitree G1 Robot Description

  Source
• dimensionalOS/dimos

  2024-10-19 - Research Publication - Dimensional is the agentic operating system for physical space. Command humanoids, quadrupeds, drones, and other hardware platforms in natural language and build multi-agent systems that work seamlessly with physical input (cameras, lidar, actuators).

  Source
• unitreerobotics/xr_teleoperate

  2024-08-05 - Research Publication - This repository implements teleoperation of the Unitree humanoid robot using XR Devices.

  Source
• fan-ziqi/robot_lab

  2024-07-06 - Research Publication - The source registers exact model-specific Isaac Lab environments for tracked quadruped, wheeled-quadruped and humanoid platforms.

  Source
• kevinzakka/mink

  2024-02-26 - Research Publication - Python inverse kinematics based on MuJoCo.

  Source
• unitreerobotics/unitree_mujoco

  Research Publication - Unitree’s MuJoCo simulator integrates low-level SDK2 control with C++ and Python simulation. It supplies robot models, joint-command/state interfaces, terrain tools and controller examples for digital validation.

  Source
• Unitree robot descriptions and simulation packages

  Research Publication - Unitree's shared robot-description and simulation repository publishes robot models, meshes and ROS control packages for multiple platforms.

  Source
• Unitree SDK2 control interfaces and examples

  Research Publication - Unitree SDK2 supplies C++ communication libraries and platform-specific examples for robot control. The source separates G1, H1, H2 and R1 examples alongside quadrupeds and shared DDS interfaces.

  Source

Media library

• G1 full-body view

  image · Unitree Robotics

  Source
• Whole-body articulation

  image · Unitree Robotics

  Source
• Platform detail

  image · Unitree Robotics

  Source
• Whole-body control demonstration

  video · Unitree Robotics

  Source
• Expanded skills demonstration

  video · Unitree Robotics

  Source

Official technical references and supporting media for Unitree G1.

• G1 manuals and downloads

  WEB / PDF · G1 · Unitree Robotics

  Source
• G1 end prosthetic hand installation guide

  PDF · Hand · Unitree Robotics

  Source

• Unitree says cumulative G1 production reaches roughly 10,000–20,000

  2026-10-22 - Production Milestone - Wang made the disclosure at the 2026 World Physical AI Conference on October22. He identified manufacturing scale, Physical AI model development and compute infrastructure as priorities for broader adoption. Unitree is developing a self-improvement workflow connecting literature search, code generation, simulation and physical testing.

  Source
• Unitree G1 performs five contact-rich interactions in REACT study

  2026-10-20 - Research Publication - A VLM-guided pipeline preserves meaningful human contact while retargeting demonstrations to the humanoid. A privileged teacher is distilled into an egocentric depth policy, with generative partner-state prediction for occlusion. Training code is planned for early November; the current repository provides project information and demonstration media.

  Source
• Imacreate launches motion-reproduction software for Unitree G1

  2026-10-19 - Product Launch - The annual licence costs JPY 5 million before tax, excluding the robot; implementation support starts at JPY 3 million per month. Processing and motion data remain on a local GPU-equipped computer. In company testing, G1 performed kendama for about an hour without intervention, with approximately 4,000 successes per hand.

  Source
• Unitree G1 tested for outdoor okra detection and grasping

  2026-09-29 - Demonstration - The configuration combined a ZED-M stereo camera, Dex1-1 hand and cutting attachment, left-arm basket and Jetson Orin 64GB computing. Grasping succeeded in 16 of 20 indoor trials. Outdoor detection and grasping were demonstrated, but the cutting tool remained unfinished and the trial did not complete reliable harvesting.

  Source
• Unitree introduces UnifoLM-WLA-1.0 humanoid foundation model

  2026-09-10 - Research Publication - The model was trained on approximately 2,500 hours of real-robot data spanning 54 tabletop and 10 whole-body tasks. A shared action representation supports two-finger grippers and multiple five-finger hands; the complete base weights and fine-tuning code followed on September 28.

  Source
• Unitree shows claimed autonomous G1 sparring

  2026-09-07 - Demonstration - The robot wore gloves and maintained bipedal balance while stepping and changing posture during contact. Unitree attributed control to UnifoLM-X2-1.0 and claimed its world model predicts near-future physical states in real time.

  Source
• Unitree introduces UnifoLM-OminiA-0.3 with humanoid task demonstration

  2026-07-20 - Demonstration - Demonstrated tasks included moving a pillow to a sofa, selecting a specified medicine box and putting clothing into a basket. The video also showed dishwasher loading and voice-controlled bed-height adjustment, including stopping on a spoken command.

  Source
• Asynchronous tracking study validates G1 sparse-motion control

  2026-06-24 - Research Publication - Physical 29-DoF G1 experiments included large squat-like motions commanded through head-and-hand trajectories. Self-guidance constrained otherwise unspecified lower-body movement; the paper reports numerical tracking comparisons in simulation alongside hardware demonstrations.

  Source
• TaskNPoint study tests G1 sports and box interaction

  2026-06-24 - Research Publication - Hardware evaluation covered tennis contact, soccer contact and box pick-and-place, with 20 trials per condition; sports success required contact rather than a completed game. An eight-camera OptiTrack system tracked objects at 120 Hz, supplying predicted contact positions and timing to the controller.

  Source
• Texedo study executes language-selected motions on G1

  2026-06-22 - Research Publication - The hardware evaluation used SONIC to track 30 selected trajectories, including dancing, walking, stepping and boxing-like motion. All 30 physical trajectories finished without falling or early termination; the study also reports joint-position, velocity and acceleration tracking errors.

  Source
• LP-NavOA study tests obstacle bypass on G1

  2026-06-22 - Research Publication - A physical G1 used odometry and local range measurements to bypass wall barriers and recover its commanded direction at 0.5–1.0 m/s. The closest observed clearance in the physical run was 0.281 m; a separate simulation benchmark covered 1,500 episodes.

  Source
• VQActFlow study benchmarks G1 task selection

  2026-06-19 - Research Publication - The G1 used Inspire RH56DFTP hands and a head-mounted RealSense D435i camera; four pick-and-place tasks each had 140 teleoperated demonstrations. Across 20 trials per task, stronger classifier-free guidance raised aggregate success from 23.8% to 57.5%, chiefly by reducing wrong-object actions.

  Source
• DUET study tests G1 collaboration with Vega1

  2026-06-18 - Research Publication - A Unitree G1 and a Dexmate Vega1 performed four collaborative tasks: sweeping, transferring, board balancing and a doll handover. The system combined paired human demonstrations with two-operator robot teleoperation and deployed a centralized action-chunking policy.

  Source
• HALOMI study learns G1 loco-manipulation from human demonstrations

  2026-06-17 - Research Publication - The physical G1 had a custom actuated neck; five tasks included bag transfer, bread placement, towel transfer, tossing and squat-and-grasp. Three manipulation tasks received 20 evaluation rollouts per setting, while tossing and squat-and-grasp were assessed through demonstrations.

  Source
• SteerGenPO study transfers stable walking to G1

  2026-06-15 - Research Publication - The physical G1 executed stable forward walking with the simulation-trained controller and no hardware fine-tuning. The paper separately evaluates six Isaac Lab benchmarks and closed-loop path following in MuJoCo.

  Source
• PTDL study tests G1 recovery on moderate terrain

  2026-06-08 - Research Publication - Hardware tests used flat ground, gravel and 5° and 10° slopes, followed by a 0.3 m/s forward command after standing. The controller used proprioception; 20° slopes belonged to simulation, and a second fall triggered a safety stop in hardware trials.

  Source
• MotionWAM study demonstrates whole-body G1 control

  2026-06-08 - Research Publication - The G1 used dual ALOHA2 grippers and a head-mounted RealSense D435i camera; a workstation policy server ran in a closed control loop. Evaluation covered 20 trials per task, including coordinated torso movement, squatting, hand manipulation and task-driven foot interaction.

  Source
• Marope study demonstrates cooperative G1 rope skipping

  2026-06-06 - Research Publication - Decentralized rope-control policies are coordinated by a higher-level scheduler and trained against varied jumping behaviors. Physical experiments used motion capture for partner state; extending collaboration to onboard sensing remained future work.

  Source
• Predictive Style Matching study tests natural G1 locomotion

  2026-06-05 - Research Publication - The predictor maps lower-body history and velocity commands to upper-body and gait targets during training, then is removed for deployment. Physical G1 tests assessed arm swing and push recovery; quantitative style and fall-rate benchmarks were also evaluated in simulation.

  Source
• State-dependent AMP study unifies G1 locomotion and recovery

  2026-05-18 - Research Publication - A frozen ONNX policy ran at 50 Hz on physical G1 hardware, recovering from prone and supine starts and transitioning between walking and running. Training used three LAFAN1 reference clips and separate recovery and locomotion discriminators; deployment required no runtime mode-switching command.

  Source
• IMU teleoperation study validates G1 whole-body motion

  2026-05-12 - Research Publication - The same processing and retargeting pipeline ran in MuJoCo and on physical G1 hardware without modification. Demonstrated movements included walking, sitting, turning, bowing and coordinated full-body gestures.

  Source
• Rapid-grasp study tests G1 vision-guided manipulation

  2026-04-19 - Research Publication - The G1 used Dex3 hands, a RealSense D435i and an external RTX 4060 laptop; bottle tests placed a 22 cm object at five positions in a 20 × 20 cm workspace. The same perception-to-motion pipeline was also demonstrated in window glue application; object-detection scores were evaluated separately from grasp completion.

  Source
• Terrain-aware motion study demonstrates G1 obstacle traversal

  2026-04-19 - Research Publication - Physical tests covered boxes, stairs and vaulting over hurdles, including mixed courses absent from reinforcement-learning fine-tuning. The robot used its knees and hands for support on climbs and hand contact to cushion descents; quantitative success comparisons were conducted in simulation.

  Source
• EgoNav study transfers human navigation data to G1

  2026-04-01 - Research Publication - The system combines visual memory, frozen video features, diffusion trajectory prediction and receding-horizon control. Physical G1 tests in unseen indoor and outdoor scenes included waiting for doors, avoiding crowds and navigating near glass walls without robot-data fine-tuning.

  Source
• Sensor-ablation study benchmarks G1 manipulation

  2026-03-30 - Research Publication - The study tested 14 sensor combinations on a G1 EDU-U4 with three-finger hands, using can sorting and cube grasping tasks. A minimal active stereo-camera setup achieved 94.4% structured-task success and 87.5% spatial-generalization success; extra modalities sometimes reduced performance in the limited-data setting.

  Source
• Chasing Autonomy study demonstrates G1 running

  2026-03-26 - Research Publication - Physical G1 tests reached 3.3 m/s and traversed routes hundreds of metres long. The controller was integrated with perception and planning for outdoor obstacle avoidance, with velocity commands guiding locomotion.

  Source
• ECHO study tests language-to-motion control on G1

  2026-03-17 - Research Publication - The physical platform was a 29-DoF G1 EDU; four gesture commands each received 20 trials, with all 80 completed under the paper’s success criterion. A compact 38-dimensional representation carried joint angles, planar root velocity, height and orientation from the cloud generator to the edge controller.

  Source
• Psi-Zero study tests G1 long-horizon manipulation

  2026-03-12 - Research Publication - The physical configuration paired a 29-DoF G1 with Dex3-1 hands and a RealSense D435i camera; MANUS gloves and PICO tracking supported data collection. Evaluation covered eight real-world tasks with three to five subtasks each, using 80 teleoperated fine-tuning demonstrations per task.

  Source
• XL-VLA study demonstrates G1 dexterous manipulation

  2026-03-10 - Research Publication - The physical G1 configuration used Inspire hands and a chest-mounted RealSense L515 camera; MANUS gloves supplied teleoperation demonstrations. Across four G1 tasks, mean success was 0.825 for XL-VLA versus 0.525 for the raw-action π0 baseline.

  Source
• ZeroWBC study demonstrates G1 control from human video

  2026-03-10 - Research Publication - Physical tasks included navigation, kicking, sofa sitting and box moving, using motion tracking after human-to-robot retargeting. Box carrying succeeded in 30% of trials; grasp slipping and small perception errors limited the physical results.

  Source
• Unitree releases whole-body teleoperation dataset

  2026-03-05 - Research Publication - The release focuses on real-robot whole-body teleoperation, combining movement and manipulation recordings. Unitree announced high-frequency rolling updates as additional data becomes available.

  Source
• ULTRA study tests multimodal G1 loco-manipulation

  2026-03-03 - Research Publication - Physical tests compared motion-capture control with onboard egocentric sensing for vertical and lateral object goals. Egocentric goal following completed 5/10 vertical and 6/10 lateral trials; motion capture supplied success evaluation in every setting.

  Source
• Rhythm study demonstrates interaction between two G1 robots

  2026-03-03 - Research Publication - Hardware demonstrations included coordinated hugging and dancing rather than two independently replayed motions. A C++ onboard stack ran policy inference at 50 Hz and joint PD control at 500 Hz, using peer-state estimates for coordination.

  Source
• Surgical-assistance study tests teleoperated G1 on cadaver

  2026-02-27 - Research Publication - An attending surgeon performed the procedure while a separate teleoperator controlled the robot-mounted camera. The proof of concept achieved stable visualization in a research operating room; the study did not involve a live patient.

  Source
• Unitree G1 robots perform martial arts at Temple of Heaven

  2026-02-23 - Demonstration - The robots wore red shirts and moved in formation through leaps, shifts and flips before a fist-clasping salute. The footage followed Unitree’s separate February 16 Spring Festival Gala appearance.

  Source
• Unitree G1 listed for sale on Amazon US

  2026-02-19 - Market Signal - The listing specified a “No Secondary Development” model controlled through a handheld remote or mobile app, rather than the programmable G1-EDU. Reported hardware included 23 degrees of freedom, a 127 cm frame and approximately two hours of battery life.

  Source
• Perceptive Humanoid Parkour study demonstrates G1 climbing

  2026-02-17 - Research Publication - A physical 29-DoF G1 climbed a 1.25 m wall in 3.63 seconds and completed a 48-second obstacle course with adaptation to displaced obstacles. Skill execution used onboard perception; navigation received simple two-dimensional velocity commands.

  Source
• APEX study demonstrates G1 high-platform traversal

  2026-02-11 - Research Publication - Physical experiments used a 29-DoF G1 to traverse 0.8 m platforms and demonstrate climbing down, standing up and lying down. Livox MID360 lidar supplied elevation maps processed on an Intel Core i7, with filtering designed to reduce perception differences between training and hardware.

  Source
• EAGLE study transfers control across four humanoids

  2026-02-03 - Research Publication - A policy trained in simulation controlled physical Unitree H1, Unitree G1, Booster T1 and Fourier N1 robots without hardware-specific fine-tuning. Hardware demonstrations covered walking, leaning and squatting; a fifth platform, PNDbotics Adam, was included in simulation.

  Source
• HUSKY study demonstrates G1 skateboard control

  2026-02-03 - Research Publication - The framework models skateboard tilt and steering geometry, then uses motion priors and trajectory guidance to switch between pushing and steering. Physical G1 trials used relatively simple terrain; onboard visual state estimation remained future work.

  Source
• HumanoidVLM study tests G1 contact-rich manipulation

  2026-01-21 - Research Publication - Physical tabletop tasks included surface following, massage-ball contact, bimanual object placement, fork use and grasping. Parameter retrieval was evaluated on 14 egocentric images, with 93% accuracy; robot experiments separately assessed stable contact behavior.

  Source
• FRoM-W1 study tests language-driven H1 and G1 control

  2026-01-19 - Research Publication - Physical deployments used H1 with Inspire hands and a torso ZED Mini camera, and a 29-DoF G1 EDU+ with Dex3-1 hands. A separate workstation ran the language model, while onboard computers executed control; the G1 waist pitch and roll joints were locked for stability.

  Source
• Unitree introduces humanoid robot App Store beta

  2025-12-12 - Product Launch - Early G1 collections included Funny Actions, Twist Dance and Bruce Lee-inspired martial-arts routines. Developers could upload their own code, and Unitree offered rewards for exceptional contributors.

  Source
• Unitree demonstrates full-body humanoid teleoperation platform

  2025-11-06 - Demonstration - Operators wearing motion-capture straps controlled soccer, martial-arts and boxing movements in real time. The demonstration also showed vacuuming, laundry handling and refrigerator loading as examples of human-guided data collection.

  Source
• Layered-control study validates G1 stairs and ledges

  2025-10-16 - Research Publication - Hardware evaluation covered stair ascent, stair descent, a hinged ledge and a soft ledge, with five trials for each policy and task. Two RealSense D435 cameras supplied perception processed on Jetson Orin NX; a separate Ryzen laptop ran the controller.

  Source
• DemoHLM study transfers simulated demonstrations to G1

  2025-10-13 - Research Publication - The training pipeline expands a single simulated demonstration into synthetic data for imitation learning. A physical G1 with an RGB-D camera completed ten task evaluations under changed spatial conditions, using closed-loop visual feedback.

  Source
• Humanoid Everyday releases G1 and H1 manipulation data

  2025-10-09 - Research Publication - The initial paper described 10,300 trajectories and more than three million frames spanning 260 tasks in seven categories. Collection used a 29-DoF G1 with Dex3-1 hands and a 27-DoF H1 configuration with Inspire hands; RGB-D, lidar and robot-state streams accompanied demonstrations.

  Source
• RobotDancing study tests long-horizon humanoid tracking

  2025-09-25 - Research Publication - Physical G1 tests completed several LAFAN1 dance sequences; three sequences were untested because of robot maintenance. H1 and H1-2 demonstrations extended the training recipe to larger hardware, where torque-to-weight limits constrained extreme movement.

  Source
• RuN study demonstrates G1 walk-run transitions

  2025-09-25 - Research Publication - Physical G1 trials transferred the simulation-trained policy without additional hardware tuning, covering locomotion from standing through running at up to 2.5 m/s. The residual policy corrects dynamic interactions while the motion generator supplies a human-like movement prior.

  Source
• RoMoCo toolbox demonstrates G1 model-based stepping

  2025-09-23 - Research Publication - Physical G1 experiments used a divergent-component-of-motion planner and position inverse kinematics to step in place, with measured centre-of-mass and foot-height tracking. Upper-body joints stayed at fixed nominal angles; H1 evaluation in the paper used simulation.

  Source
• HDMI study transfers human-video interaction skills to G1

  2025-09-20 - Research Publication - Physical tests included 67 consecutive door openings and traversals, plus box handling and other contact-rich tasks. Motion-capture markers on the robot pelvis and manipulated objects supplied pose measurements; the policies were trained in Isaac Sim before transfer.

  Source
• Unitree G1 demonstrates Auki-guided navigation at WOW Summit

  2025-09-09 - Demonstration - The Auki domain was prepared in a few hours the preceding day and was also accessible to visitors through app-free augmented-reality navigation. The robot combined onboard sensing with the domain’s navigable areas, obstacles, routes and booth points of interest.

  Source
• Unitree G1 interacts with fighters at UFC Performance Institute

  2025-08-28 - Demonstration - Motion data collected from UFC athletes covered boxing, Muay Thai and jiu-jitsu movements. The optical recordings captured limb-joint angles, punching speed and continuous motion trajectories. Unitree used real-time motion data to assess simulated training and refine control algorithms.

  Source
• UniTracker study validates G1 whole-body motion tracking

  2025-07-10 - Research Publication - Physical G1 experiments tracked walking, kicking, turning, dancing and squatting with one policy. The study also tested text-conditioned motion generation and video-based motion estimates as previously unseen motion inputs.

  Source
• NVIDIA posts G1 fruit teleoperation dataset

  2025-06-10 - Research Publication - Recordings use a RealSense camera and Unitree tri-finger hands, with prompts covering apple, pear, starfruit and grape handling. The CC BY 4.0 release includes RGB video and 43-dimensional joint-state and action vectors.

  Source
• Gait-conditioned learning study transfers walking to G1

  2025-05-27 - Research Publication - Physical validation covered standing, walking and walk-to-stand transitions with coordinated arm and leg motion. Running remained a simulation result; hardware transfer used the trained policy without fine-tuning.

  Source
• Unitree G1 performs a standing side flip

  2025-03-31 - Demonstration - The robot landed primarily on its left leg, then stabilized as the other foot contacted the floor. Unitree told Live Science it mainly upgraded the control algorithm and trained the maneuver using reinforcement learning.

  Source
• Unitree G1 demonstrates kip-up and balance recovery

  2025-03-21 - Demonstration - The same footage showed Tai Chi-inspired movements alongside a human partner. G1 regained its balance after a kick from behind; Unitree described the footage as normal-speed recording without AI-generated enhancement.

  Source
• Unitree G1 dance demonstration uses Westlake Robotics motion model

  2025-03-21 - Demonstration - The footage included single-leg spins, two aerial kicks, a stable finish and a final bow. Westlake said the sequence ran from one end-to-end model without remote control or accelerated playback. The company also described Charleston dancing and said the model could control multiple mass-produced robot platforms.

  Source
• HoST study demonstrates G1 standing from varied postures

  2025-02-12 - Research Publication - Physical tests used a 23-actuator G1 with an onboard Jetson Orin NX and IMU and joint-encoder feedback. The policies transferred directly from simulation to laboratory and outdoor settings, with smoothness and motion-speed constraints to reduce violent movements.

  Source
• ASAP study improves agile G1 motion transfer

  2025-02-03 - Research Publication - The hardware comparison tested kicking and an out-of-distribution LeBron James “Silencer” motion, measuring closed-loop tracking error against an untuned baseline. The residual model was used during simulation fine-tuning and removed for physical execution; a locomotion policy balanced the robot between skills.

  Source
• Unitree reports G1 gait improvement after two days of training

  2024-11-01 - Demonstration - Unitree described combining simulation with real-world testing for robot training. The footage also showed the company’s Go2 balancing on two legs while being pushed and kicked.

  Source
• Unitree launches G1 humanoid robot

  2024-05-13 - Product Launch - The approximately 127 cm, 35 kg platform includes Intel RealSense D435 sensing and LIVOX MID360 lidar, plus a quick-release battery rated for about two hours. G1 EDU supports optional Dex3-1 hands, tactile sensing and NVIDIA Jetson Orin computing; its configuration and price depend on customer requirements.

  Source

Market context

Compact humanoid platform with 43 DoF, 3D LiDAR and depth sensing.

• Target environment: Research labs, university robotics programs and developer workshops.
• Workflow context: Experimentation, locomotion research, teleoperation and embodied AI prototyping at a lower cost barrier than full-size humanoids.
• Customer context: Robotics researchers, AI labs and developers seeking an accessible humanoid platform.
• Deployment model: Commercial product sold as a development platform, not as a workplace automation tool.
• Commercial maturity: Commercial product with broad developer adoption due to its low entry price.
• Adoption constraints: Limited payload and reach restrict it to research and demonstration workflows rather than real workplace tasks.
• Market position: Sits at the affordable end of the humanoid spectrum. Expands access to humanoid development but is not a deployment-ready industrial platform.
• Adjacent products: H1, AI Sapiens K0

Comparable Systems

• AI Sapiens K0

  ROBOTIS - humanoid robotics

• Booster T1

  Booster Robotics - humanoid robotics

• Booster T2

  Booster Robotics - humanoid robotics

• G1+

  Unitree - humanoid robotics

• H1

  Unitree - humanoid robotics

• LimX Oli

  LimX Dynamics - Humanoid Platform

Related Systems

• Digit

  Agility Robotics - humanoid robotics