Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces KiRAS, a keyframe-guided self-imitation learning framework for quadruped robots to acquire diverse skills and adapt to complex terrains. KiRAS uses keyframes as skill representations to learn diverse skills on flat terrain through self-imitation, followed by continued training on rough terrains to enhance robustness, employing a proficiency-based skill initialization technique to mitigate catastrophic forgetting. Experiments on Solo-8 and Unitree Go1 robots demonstrate KiRAS enables robust skill acquisition and smooth transitions across challenging terrains, offering a lightweight platform for multi-skill generation.
Quadruped robots can now learn diverse skills and adapt to complex terrains without expert datasets, thanks to a novel keyframe-guided self-imitation learning framework.
With advances in reinforcement learning and imitation learning, quadruped robots can acquire diverse skills within a single policy by imitating multiple skill-specific datasets. However, the lack of datasets on complex terrains limits the ability of such multi-skill policies to generalize effectively in unstructured environments. Inspired by animation, we adopt keyframes as minimal and universal skill representations, relaxing dataset constraints and enabling the integration of terrain adaptability with skill diversity. We propose Keyframe Guided Self-Imitation for Robust and Adaptive Skill Learning (KiRAS), an end-to-end framework for acquiring and transitioning between diverse skill primitives on complex terrains. KiRAS first learns diverse skills on flat terrain through keyframe-guided self-imitation, eliminating the need for expert datasets; then continues training the same policy network on rough terrains to enhance robustness. To eliminate catastrophic forgetting, a proficiency-based Skill Initialization Technique is introduced. Experiments on Solo-8 and Unitree Go1 robots show that KiRAS enables robust skill acquisition and smooth transitions across challenging terrains. This framework demonstrates its potential as a lightweight platform for multi-skill generation and dataset collection. It further enables flexible skill transitions that enhance locomotion on challenging terrains.