Event
Differentiable Contact Dynamics Gradients for Legged Locomotion
1 reports · 1 sources
Gijeong Kim and co-authors propose a smooth analytical gradient for frictional contact that replaces only the exact contact sensitivity in backpropagation, while the forward hard-contact solve and physically feasible trajectories remain unchanged. The authors claim that a policy trained on a desktop in under 2 minutes using 100,000 samples can be transferred to hardware without fine-tuning; under a trotting policy, the robot can withstand pushes and traverse rubble and slippery ground. In trajectory optimization, the authors claim that this gradient can find contact-rich solutions that exact gradients and Signorini relaxation gradients miss.
Report timeline
- Differentiable Contact Dynamics for Legged Locomotion: Policy Learning and Contact-Implicit Trajectory Optimization
This paper proposes a smooth analytic gradient for frictional contact that replaces the exact contact sensitivity only in the backward pass, leaving the forward hard-contact solve and the hardware-feasible trajectory unchanged. A policy trained on 100,000 samples in under 2 minutes on a desktop PC transfers to hardware without fine-tuning; under a trotting policy the robot withstands pushes and traverses gravel and slippery ground. In trajectory optimization, the gradient finds contact-rich solutions that the exact gradient and the Signorini relaxation gradient miss.
arXiv cs.RO ·