E2DRL: End-to-End Differentiable Reinforcement Learning for 3D Obstacle Avoidance in Quadruped Robots
E2DRL: End-to-End Differentiable Reinforcement Learning for 3D Obstacle Avoidance in Quadruped Robots
This research proposes E2DRL, a differentiable reinforcement learning framework for 3D obstacle avoidance in quadruped robots. The method uses a soft obstacle-interaction formulation with velocity-conditioned stop-gradient shaping to provide first-order gradients for training. A privileged teacher trained via SAPO is distilled into a depth-based student using DAgger, and integrated with a terrain-aware gating module for deployment. The approach achieves higher sample efficiency and better performance in simulation compared to PPO and DCBF NMPC, with the first sim-to-real validation on a Unitree Go2 robot.
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.
- Environment
- SimulationOpen source S4
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
n a differentiable simulator is distilled into a recurrent depth-based student via DAgger, then blended with a base locomotion policy using a terrain-aware gating module for deployment. In simulation, E2DRL achieves higher sample efficiency than PPO and demonstrates superior velocity-tracking performance and 3D obstacle-avoidance capability compared with both PPO and Duality-based DCBF NMPC. We present the first sim-
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org