FACT: A Multimodal Force-Action-Track Foundation Model for Human-to-Robot Dexterity - 2026-10-13 - Research Publication - FACT combines visual planning with force-aware action refinement for dexterous manipulation. Its ten-task physical evaluation uses YAM Ultra arms and Sharpa Wave hands; robot pretraining also reuses the T-Rex collection… - source: arxiv.org
What Should Embodied Agents Remember? Benchmarking and Evaluating Memory Modality for Embodied Agents - 2026-10-08 - Research Publication - MeMoREval compares memory modalities for embodied agents across 53 tasks and four simulated indoor scenes, measuring task success and memory inference costs. Two physical Dexmate Vega demonstrations test how stored… - source: arxiv.org
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… - source: arxiv.org
DexTacWAM: A Visuo-Tactile World-Action Model for Dexterous Manipulation - 2026-09-21 - Research Publication - DexTacWAM extends a video world-action model with per-fingertip tactile prediction and a hand-aware tactile compressor. Across six contact-rich tasks on a bimanual Dexmate and Sharpa setup, it averages 70.6 against 38.0… - source: arxiv.org
ReVeal: A Reconstruction-Aware Real-to-Sim Framework for VLA Policy Evaluation - 2026-09-20 - Research Publication - ReVeal assesses reconstructed environments for vision-language-action policy evaluation using visual and planar geometry metrics. Eight matched physical and simulated Dexmate Vega tasks test how reconstruction fidelity… - source: arxiv.org
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… - source: arxiv.org
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… - source: arxiv.org
T-Rex: Tactile-Reactive Dexterous Manipulation - 2026-06-15 - Research Publication - T-Rex evaluates two physical 22-DoF Sharpa Wave hands with per-fingertip tactile deformation and six-axis wrench sensing. The work introduces 100 hours of tactile-rich robot data and reports more than 30% higher average… - source: arxiv.org
Contrastive Action-Image Pre-training for Visuomotor Control - 2026-06-15 - Research Publication - CAIP learns action-centred visual representations from 32,041 hours of egocentric human video and 88 hours of robot manipulation data. Physical evaluation uses a Dexmate Vega with two 22-DoF Sharpa Wave hands and reports… - source: arxiv.org
FTP-1: A Generalist Foundation Tactile Policy Across Tactile Sensors for Contact-Rich Manipulation - 2026-06-11 - Research Publication - FTP-1 trains a generalist tactile policy across heterogeneous sensors and embodiments using about 3,000 hours from 26 data sources. Physical evaluations include Sharpa North and a Sharpa-equipped Dexmate platform… - source: arxiv.org
PGDG: Physically Grounded Data Generation for Robust Bimanual Policy Learning from a Single Demonstration - 2026-05-20 - Research Publication - PGDG expands one bimanual demonstration into physically grounded recovery data, combining rollout sampling, diversity curation and corrective action relabeling. Four tasks test simulated learning and zero-shot transfer to… - source: arxiv.org
Residual Off-Policy RL for Finetuning Behavior Cloning Policies - 2025-09-23 - Research Publication - ResFiT improves a frozen behavior-cloning policy with lightweight residual corrections learned by off-policy reinforcement learning. Its physical experiments use a Dexmate Vega with OyMotion hands for ball picking and… - source: arxiv.org
Dexmate Vega robot-description and simulation models - Research Publication - Dexmate's model collection provides Vega robot descriptions, collision models and named hand/gripper configurations for simulation and planning. - source: github.com