DiffContact: Differentiable Contact Dynamics for Policy Learning and Contact-Implicit Trajectory Optimization
DiffContact: Differentiable Contact Dynamics for Policy Learning and Contact-Implicit Trajectory Optimization
This research presents DiffContact, a method providing smooth analytic gradients for frictional contact to facilitate policy learning and contact-implicit trajectory optimization in legged locomotion. By employing a central-path relaxation of the de Saxcé-corrected second-order-cone complementarity problem, the approach ensures continuous sensitivity through contact-mode transitions. The authors demonstrate that training with 100,000 samples on a desktop PC yields policies that transfer directly to hardware, where the robot successfully withstands pushes and traverses challenging terrains like debris and slippery ground. Additionally, the method achieves contact-rich solutions in trajectory optimization that were previously missed by exact or Signorini-relaxed gradients.
The paper introduces a differentiable contact dynamics method that enables policy learning and trajectory optimization for legged robots, demonstrating hardware transfer without fine-tuning on tasks involving pushes, debris, and slippery ground.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
training_samples
100,000
View original evidence
th obtained by a Newton iteration initialized from the hard-contact solution. The sensitivity remains continuous through contact-mode transitions and converges to the exact sensitivity as the relaxation vanishes. Using 100,000 samples and less than 2 min of training on a desktop PC, policy learning with the proposed gradient produces a policy that transfers to hardware without fine-tuning. On hardware, the robot unde
Open source S4Reported duration
<2 minutes
View original evidence
th obtained by a Newton iteration initialized from the hard-contact solution. The sensitivity remains continuous through contact-mode transitions and converges to the exact sensitivity as the relaxation vanishes. Using 100,000 samples and less than 2 min of training on a desktop PC, policy learning with the proposed gradient produces a policy that transfers to hardware without fine-tuning. On hardware, the robot unde
Open source S4
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
th obtained by a Newton iteration initialized from the hard-contact solution. The sensitivity remains continuous through contact-mode transitions and converges to the exact sensitivity as the relaxation vanishes. Using 100,000 samples and less than 2 min of training on a desktop PC, policy learning with the proposed gradient produces a policy that transfers to hardware without fine-tuning. On hardware, the robot unde
Open source S4
r a trotting policy withstands pushes and traverses debris and slippery ground. In trajectory optimization, the proposed gradient reaches contact-rich solutions missed by exact and Signorini-relaxed gradients. Project page: https://diffcontact.github.io
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org