SoloParkour: Constrained Reinforcement Learning for Visual Locomotion from Privileged Experience
This addresses the problem of enabling legged robots to navigate complex environments safely and efficiently for robotics applications, representing a strong specific gain but not a new paradigm.
The paper tackled the challenge of training quadruped robots to perform agile parkour skills like walking, climbing, leaping, and crawling from depth images, by introducing a constrained reinforcement learning method that uses privileged experience to warm-start visual policies, resulting in successful real-world demonstrations on a Solo-12 robot.
Parkour poses a significant challenge for legged robots, requiring navigation through complex environments with agility and precision based on limited sensory inputs. In this work, we introduce a novel method for training end-to-end visual policies, from depth pixels to robot control commands, to achieve agile and safe quadruped locomotion. We formulate robot parkour as a constrained reinforcement learning (RL) problem designed to maximize the emergence of agile skills within the robot's physical limits while ensuring safety. We first train a policy without vision using privileged information about the robot's surroundings. We then generate experience from this privileged policy to warm-start a sample efficient off-policy RL algorithm from depth images. This allows the robot to adapt behaviors from this privileged experience to visual locomotion while circumventing the high computational costs of RL directly from pixels. We demonstrate the effectiveness of our method on a real Solo-12 robot, showcasing its capability to perform a variety of parkour skills such as walking, climbing, leaping, and crawling.