To traverse complex terrains, quadrupedal robots require seamless integration of multiple motor skills, smooth gait transitions, and high-speed perceptive locomotion using only onboard sensors. We present APT-RL (Action Pretrained Transformer-based Reinforcement Learning), a unified framework that enables multi-skill locomotion for high-speed traversal in complex environments via autonomous skill transitions relying solely on onboard perception and computation.
Our approach generates large-scale, feature-rich 2D motion datasets through trajectory optimization with simplified dynamics. These datasets facilitate the training of diverse, reusable locomotion skills that effectively transfer to real quadruped robots operating on complex uneven terrains. The resulting high-quality skills serve as strong priors for efficient learning of complex downstream tasks and extend naturally to 3D environments, enabling smooth, high-speed multi-skill locomotion in deployed policy.
Real-world experiments demonstrate the framework's capabilities: the robot performs agile maneuvers through complex indoor obstacles and outdoor wild environments, including dynamic drop-down maneuvers that reach instantaneous peak speeds of up to 6 meters per second. A single onboard policy enables robust traversal of diverse obstacles, including stairs, hurdles, stepping stones, gaps, and fallen branches, showcasing the versatility and effectiveness of our approach.
Blogger's Review: This study showcases the locomotion capabilities of quadrupedal robots in complex environments, particularly through the multi-skill transitions enabled by the APT-RL framework, indicating a promising future for robotic technologies. The ability to adapt to complex terrains is crucial for achieving efficient autonomous operation in the real world, highlighting significant research value and application potential.