Skip to content

Event

∇Mimic: Differentiable Simulation Accelerates Humanoid Whole-Body Control

1 reports · 1 sources

Weyl Lu and Yubei Chen have released ∇Mimic, a humanoid whole-body control method, together with the accompanying NablaEngine differentiable GPU simulation engine; both are released as open source and are intended for first-order humanoid policy learning. NablaEngine is built for the Unitree G1, and the authors state that it eliminates simulation cost. The authors report that when tracking all 40 LAFAN1 motion segments with a single policy on a single GPU, ∇Mimic reaches a score of 0.82 in 20 minutes, whereas the world-model method takes 1.2 hours to reach the same score; after 1.8 hours, ∇Mimic rises to a score of 0.874. The above comparison and scores are all results self-reported by the authors.

Report timeline

  1. ∇Mimic: Efficient Whole-Body Control for Humanoid Robots with Differentiable Simulation

    The authors propose ∇Mimic, which uses exact gradients from a differentiable simulator to train whole-body motion-tracking policies for humanoid robots, reaching a score of 0.82 in 20 minutes and 0.874 in 1.8 hours on a single GPU, whereas the world-model method takes 1.2 hours to reach 0.82.

    arXiv cs.RO ·

Related topics

Unitree RoboticsMotion controlRobot models and learningData and simulationResearch papers

Interest trend

Current
3
Peak
4Nov 4 01:00
Past 24 hours
–

The trend compares participants observed throughout the same period, so its range may be smaller than the current interest count. Move or tap to inspect each hour; use the arrow keys on a keyboard.