Reproduce “PRIME: Physically-consistent Robotic Inertial and Motion Estimation for Legged and Humanoid Robots” 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

Jiarong Kang, Kunzhao Ren, Tao Pang, Xiaobin Xiong

The PRIME framework simultaneously estimates robot motion, contact forces, and uncertain inertial parameters for legged and humanoid robots by integrating differentiable contact dynamics into an optimization-based approach. This enables the reconstruction of physically consistent trajectories from noisy sensor data, enhancing model accuracy for downstream control and learning tasks.

Key insights

  • Explicitly modeling differentiable contact dynamics allows for robust optimization across diverse and frequent contact transitions, providing well-behaved gradients for parameter estimation.
  • Jointly estimating motion, contact forces, and inertial parameters within a single framework ensures full physical consistency, unlike decoupled approaches.
  • Utilizing Log-Cholesky parameterization for inertial properties guarantees their physical feasibility and provides smooth derivatives for gradient-based optimization.

Summary by alphaXiv.

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