Reproduce “Whole-Body Planning for Humanoids Navigating Confined Spaces via Self-Collision Avoidance References” 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
Carlos Gonzalez, Luis Sentis
Researchers from The University of Texas at Austin developed a whole-body planning framework for humanoid robots to navigate highly confined spaces, leveraging differentiable self-collision avoidance and morphology-aware guides. The system successfully generated 12-18 second dynamically feasible plans for a Unitree G1 humanoid in environments exceeding NIST emergency response standards, achieving up to 100% success where baseline planners failed, and enabled training a robust real-time policy with over 95% traversal success in simulation.
Key insights
- Directly accounting for robot rigid-body volumes through morphology-aware guides is essential for generating feasible whole-body plans in highly confined environments.
- Differentiable collision detection, integrated into a kinematic planning stage, effectively refines initial paths to avoid self-collisions without relying on simplified bounding volumes.
- A sequential optimization pipeline, including an initial dynamics feasibility pass and warm-starts, significantly improves the numerical stability and convergence of complex whole-body trajectory optimization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.