Reproduce “ADP: Adversarial Dynamics Priors for Physically Grounded Humanoid Locomotion” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Seokju Lee, Jeongtae Lee, Jeonghyeok Lim, Jeonguk Kang, Byungwook Lee, Seungho Han, Keun Ha Choi, Dongil Park and 1 more

ADP trains a humanoid locomotion policy with rewards for temporal windows of center-of-mass motion, momentum, foot forces, and contacts, rather than pose matching. Single-rigid-body dynamics optimization supplies reference trajectories, and an adversarial discriminator compares these features. In simulation, ADP reaches J80 of 115.5 N·s versus AMP’s 99.0 N·s, but hardware evidence is qualitative.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.