Unitree Research

Research

  • Noitom releases the HiPHI motion and interaction dataset - 2026-08-19 - Research Publication - The dataset contains recordings from 132 performers, including 371.8 hours of motion and 245.7 hours of interaction data. Noitom reports 90 Hz capture across 40 objects in 12 categories. Policies trained with HiPHI were… - source: globenewswire.com
  • Learning Asynchronous Upper-body Task-space Trajectory Tracking Policy for Humanoid Robots - 2026-06-24 - Research Publication - The paper “Learning Asynchronous Upper-body Task-space Trajectory Tracking Policy for Humanoid Robots” was published on June 24, 2026. The arXiv paper studies asynchronous upper-body task-space tracking on Unitree G1. The… - source: arxiv.org
  • TaskNPoint: How to Teach Your Humanoid to Hit a Backhand in Minutes - 2026-06-24 - Research Publication - The paper “TaskNPoint: How to Teach Your Humanoid to Hit a Backhand in Minutes” was published on June 24, 2026. The arXiv paper teaches dynamic backhand-hitting skills on a Unitree G1 humanoid. The study’s named evaluation… - source: arxiv.org
  • StairMaster paper deploys hollow-stair locomotion policy on Unitree Go2 - 2026-06-24 - Research Publication - StairMaster is a three-stage reinforcement-learning framework for quadruped locomotion on risky hollow stairs. The paper deploys the policy on a Unitree Go2 robot and reports zero-shot stair climbing up to a 55-degree… - source: arxiv.org
  • RoboAtlas paper evaluates contextual active SLAM on Unitree Go2 - 2026-06-24 - Research Publication - RoboAtlas combines frontier exploration, global semantic-map reasoning, and egocentric VLM reasoning for contextual Active SLAM. The system is evaluated on a Unitree Go2 in real-world environments exceeding 1,800 square… - source: arxiv.org
  • Wheeled-quadruped racing paper controls load transfer on Unitree Go2-W - 2026-06-24 - Research Publication - The paper presents a hierarchical MPC and reinforcement-learning framework for active roll control during autonomous wheeled-quadruped racing. The controller uses a vehicle-dynamics bicycle model of the Unitree Go2-W… - source: arxiv.org
  • fARfetch: Enabling Collocated AR-HRC in Large Visually Diverse Environments with VLM-Driven AR Content Adaptation - 2026-06-23 - Research Publication - The arXiv paper uses Unitree Go2 for collocated augmented-reality human-robot collaboration in visually diverse environments. The study’s named evaluation platform is Go2 from Unitree. - source: arxiv.org
  • RGB: RL Guided Whole-Body MPPI for Humanoid Control - 2026-06-23 - Research Publication - The arXiv paper combines reinforcement learning and MPPI for whole-body humanoid control on Unitree G1. The study’s named evaluation platform is G1 from Unitree. - source: arxiv.org
  • TEXEDO: Test Time Scaling for Controller-aware Language-conditioned Humanoid Motion Generation - 2026-06-22 - Research Publication - The paper “TEXEDO: Test Time Scaling for Controller-aware Language-conditioned Humanoid Motion Generation” was published on June 22, 2026. The arXiv paper studies controller-aware language-conditioned humanoid motion… - source: arxiv.org
  • LP-NavOA: Integrated Local Navigation and Obstacle Avoidance for Humanoid Robots under Limited Perception - 2026-06-22 - Research Publication - The arXiv paper evaluates local navigation and obstacle avoidance on Unitree G1 under limited perception. The study’s named evaluation platform is G1 from Unitree. - source: arxiv.org
  • CoorDex: Coordinating Body and Hand Priors for Continuous Dexterous Humanoid Loco-Manipulation - 2026-06-22 - Research Publication - The arXiv paper coordinates body and hand priors for continuous dexterous loco-manipulation on Unitree G1. The study’s named evaluation platform is G1 from Unitree. - source: arxiv.org
  • VQActFlow: Vector-Quantized Action Mode Steering for Multi-Task Robot Manipulation - 2026-06-19 - Research Publication - The paper “VQActFlow: Vector-Quantized Action Mode Steering for Multi-Task Robot Manipulation” was published on June 19, 2026. The arXiv paper studies vector-quantized action-mode steering for multi-task robot manipulation… - source: arxiv.org
  • Long-Distance Real-World Navigation of the Legged-Wheeled Robot Go2-W Using Deep Reinforcement Learning - 2026-06-19 - Research Publication - The paper builds a deep-reinforcement-learning locomotion controller and autonomous navigation system for the commercially available Go2-W. The controller extends a proprioception-only quadruped policy to the 16-DoF… - source: arxiv.org
  • Duet: Dual-Robot Understanding via Efficient Teaching - 2026-06-18 - Research Publication - The paper “Duet: Dual-Robot Understanding via Efficient Teaching” was published on June 18, 2026. The arXiv paper studies dual-robot understanding via efficient teaching with Unitree G1. The study’s named evaluation… - source: arxiv.org
  • ForEnt dataset paper collects forest-entrapment data with Unitree Go2 - 2026-06-18 - Research Publication - ForEnt is a multi-modal dataset for studying quadruped robot entrapments in forest environments. The dataset is collected with a Unitree Go2 quadruped across eight forest sites to characterize vegetation-related failure… - source: arxiv.org
  • HALOMI: Learning Humanoid Loco-Manipulation with Active Perception from Human Demonstrations - 2026-06-17 - Research Publication - The arXiv paper learns humanoid loco-manipulation with active perception from human demonstrations on Unitree G1. The study’s named evaluation platform is G1 from Unitree. - source: arxiv.org
  • Steering Generative Reinforcement Learning into Stable Robotic Controller - 2026-06-15 - Research Publication - The paper “Steering Generative Reinforcement Learning into Stable Robotic Controller” was published on June 15, 2026. The arXiv paper studies stable generative reinforcement-learning controllers on Unitree G1. The study’s… - source: arxiv.org
  • SemGeoNav paper evaluates safety-guided visual navigation on Unitree Go2 - 2026-06-15 - Research Publication - SemGeoNav integrates semantic reasoning with geometric planning for safer visual navigation in open environments. The approach is evaluated on a Unitree Go2 quadruped robot in real-world navigation experiments. - source: arxiv.org
  • FlashNav paper trains deployable navigation policies for Unitree Go2 within seconds - 2026-06-14 - Research Publication - FlashNav is a GPU-first framework for ultra-fast range-based robot navigation policy training. Experiments include TurtleBot2 and Unitree Go2, with deployable policies trained in seconds-level wall-clock time. - source: arxiv.org
  • LoComposition paper deploys terrain-adaptive quadruped locomotion on Unitree Go2 - 2026-06-14 - Research Publication - LoComposition decomposes quadruped locomotion into task rewards, operational constraints, energy minimization, and terrain-adaptive exteroceptive perception. The paper reports robust zero-shot deployment on a physical… - source: arxiv.org
  • Kine2Go paper releases kinematic dataset for Unitree Go2 gaits and motions - 2026-06-12 - Research Publication - Kine2Go introduces a kinematic dataset for the Unitree Go2 robot covering diverse gaits and motions. The dataset contains 800 gait-kinematics trajectory records derived from 40 distinct policies and translated to a… - source: arxiv.org
  • AgniNav paper transfers monocular local navigation to Unitree Go2 - 2026-06-09 - Research Publication - Researchers published AgniNav and documented cross-embodiment monocular navigation experiments that included a Unitree Go2. AgniNav standardizes cross-embodiment monocular local navigation through a configurable… - source: arxiv.org
  • PTDL:Multi-Terrain Fall Recovery via Phase-Terrain Decoupled Learning - 2026-06-08 - Research Publication - Humanoid robots can fall on slopes, gravel, and uneven ground in unstructured environments. We target integrated fall recovery and locomotion: rebuilding balance from a fallen state using proprioception alone and resuming…… - source: arxiv.org
  • 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… - source: arxiv.org
  • Semantic-skill discovery paper validates multi-quadruped coordination on Unitree Go2 teams - 2026-06-06 - Research Publication - Researchers published a semantic-skill discovery method and validated coordinated rollouts with teams of Unitree Go2 quadrupeds. The paper studies continual quadruped robot coordination through semantic skill discovery.… - source: arxiv.org
  • 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. Experiments use G1 as the… - source: arxiv.org
  • 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. Experiments use G1 as the physical… - source: arxiv.org
  • Library service-robot navigation paper builds ROS 2 stack on Unitree Go2 Edu - 2026-06-02 - Research Publication - The paper presents a ROS 2 autonomous-navigation system for a Unitree Go2 Edu quadruped operating in library aisles and cluttered reader spaces. The stack combines RTAB-Map, AMCL, EKF sensor fusion, Nav2 planning, A*, and… - source: arxiv.org
  • Hierarchical object-representation paper evaluates perception with Unitree B2 dataset - 2026-06-01 - Research Publication - The paper studies hierarchical object representation for spatial robot perception using points, meshes, and superquadrics. It evaluates perception on a dataset collected with a Unitree B2 robot platform. - source: arxiv.org
  • State-dependent AMP paper unifies G1 walking, running, and fall recovery - 2026-05-18 - Research Publication - The paper introduces a state-dependent adversarial motion prior framework that lets one policy handle walking, running, and fall recovery on Unitree G1. Hardware experiments validate a single frozen ONNX policy running at… - source: arxiv.org

Articles

View Unitree overview