AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

Cross-Embodied Co-Design for Dexterous Hands

Kehlani Fay, Darin Anthony Djapri, Anya Zorin, James Clinton, Ali El Lahib, Hao Su, Michael T. Tolley, Sha Yi, Xiaolong Wang

ICLR 2026 · Reinforcement learning › Visual and representation RL · approach: Dexterous manipulation transfer

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add fine-tuning of the best design with domain randomization (ablation)co-designed hand · In-hand rotation (sim)continuous angular velocity (rad/s): 1.85 -> 3.3Improved
use RL cross-embodied control instead of MPPI (ablation)co-design framework · In-hand rotation (sim)continuous angular velocity (rad/s): 0.62 -> 3.3Improved
use graph heuristic search with learned value network instead of RoboGrammardesign search · In-hand rotation (sim)continuous angular velocity (rad/s): 0.26 -> 3.3Improved
use graph heuristic search instead of Monte Carlo tree searchdesign search · In-hand rotation (sim)time to fall: 2.71 seconds -> no fall over three minutesImproved
use co-designed 3-finger radially symmetric hand instead of anthropomorphic handblind in-hand rotation · In-hand rotation, blind policy (sim)angular velocity (rad/s): 0.73 -> 1.85Improved
use co-designed 3-finger hand instead of blind Leap handblind in-hand rotation · In-hand rotation (sim)continuous angular velocity (rad/s): 0.0 -> 3.3Improved
use cross-embodied policy instead of individual PPO per morphologydesign evaluation · 5-finger symmetric hand rotationangular velocity (rad/s): 0.56 rad/s -> 65% improvementImproved
use cross-embodied evaluation instead of individual PPO per designdesign evaluation runtime · Co-design evaluationruntime: 20 designs at 47 minutes each -> 2,000 designs in 5.18 hours (400x speedup)Improved
use thin fingertips instead of standard fingertips (ablation)in-hand rotation hands · In-hand rotation (sim)rotation stabilityHurt
increase finger body length scale (ablation)LEAP hand parameters · Isaac-Reorient-Cube-Leapcorrelation with rotation reward: r=0.748Improved
increase palm width scale (ablation)LEAP hand parameters · Isaac-Reorient-Cube-Leapcorrelation with rotation reward: r=-0.729Hurt
increase actuator damping (ablation)LEAP hand parameters · Isaac-Reorient-Cube-Leapcorrelation with rotation reward: r=-0.476Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy