Reproduce “Impact Invariant Control with Applications to Bipedal 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
William Yang, Michael Posa
The GRASP Laboratory at the University of Pennsylvania introduced a control framework for legged robots that projects control objectives onto an "impact invariant subspace." This technique reduces control effort spikes and enhances robustness during impact events, demonstrating improved performance in simulations of walking and jumping, and smoother motor commands on a physical Cassie robot.
Key insights
- The key insight is that certain generalized velocities remain continuous through an impact event, defining an "impact invariant subspace."
- Projecting feedback control objectives onto this subspace prevents spikes from instantaneous velocity changes, enhancing robustness without sacrificing full control authority in unaffected dimensions.
- The method's generalizability allows its application to both joint space and operational space control architectures for legged robots.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.