Whole-Body Planning for Humanoids Navigating Confined Spaces via Self-Collision Avoidance References

2026-08-10

AI Summary

This work addresses the challenge of whole-body motion planning for humanoid robots in extremely confined environments, where dense obstacles, self-collision constraints, and multi-contact dynamic feasibility must be jointly satisfied. The authors propose a three-stage framework: first, a kinematic path search based on rigid-body reachable volumes is combined with differentiable collision avoidance to generate volume-aware guiding trajectories; second, these trajectories initialize a high-order trajectory optimization; and third, a residual reinforcement learning policy is trained for robust online execution. By integrating reachability constraints with differentiable collision avoidance, the method mitigates the local minima commonly encountered by conventional spline-based parameterizations in narrow spaces. Evaluated on the Unitree G1 platform, the system successfully completes tasks in passages with clearance ratios below 1.5, generating complex hand–foot multi-contact trajectories within 12–18 seconds, significantly outperforming existing baselines while maintaining high tracking accuracy under strong physical perturbations.

Abstract

Humanoid locomotion in highly confined environments requires navigating dense environmental obstacles and complex self-collision bounds while maintaining multi-contact dynamic feasibility. Traditional trajectory optimizers frequently struggle in these restricted spaces, as navigating the large collision space with splines on particle abstractions is insufficient and leads to poor local minima. To address this, we propose a three-stage whole-body planning framework that formulates kinematic path planning directly over kinematically reachable rigid-body volumes. By integrating differentiable collision avoidance into a reachability-constrained formulation, our framework synthesizes volume-informed guides that reliably guide a full-order trajectory optimizer over long horizons. We show that these optimized plans serve as high-quality references to train a residual reinforcement learning policy for robust online execution. We validate our approach on the Unitree G1 humanoid across three benchmark testbeds exceeding NIST emergency response standards, achieving restricted confinement ratios ($C_r < 1.5$). Our framework generates feasible trajectories across 12-to-18-second tasks with complex foot and hand contacts where standard baselines fail, while the learned policy successfully tracks these plans under extensive domain randomization in physics simulation.

Problem

• humanoid locomotion
• confined spaces
• self-collision avoidance
• multi-contact dynamics
• trajectory optimization

Innovation

• whole-body planning
• differentiable collision avoidance
• kinematic reachability
• residual reinforcement learning
• humanoid locomotion