KiRAS: Keyframe Guided Self-Imitation for Robust and Adaptive Skill Learning in Quadruped Robots
Xiaoyi Wei, Peng Zhai, Jiaxin Tu, Yueqi ZHang, Yuqi LI, Zonghao ZHang, Hu Zhou, Lihua ZHang
AI summary
Problem
Current multi-skill learning methods for legged robots rely heavily on flat-terrain expert datasets or complex reward engineering, which limits generalization to unstructured environments and hinders flexible task switching.
Approach
The framework uses keyframes as minimal skill targets to guide self-imitation learning on flat terrain, then fine-tunes the same policy on rough terrains while employing a proficiency-based initialization technique to prevent catastrophic forgetting.
Key results
- Keyframe-guided self-imitation eliminates expert dataset requirements for skill acquisition
- Proficiency-driven initialization balances skill learning and prevents catastrophic forgetting
- Robust multi-skill traversal and smooth transitions demonstrated on Solo-8 and Unitree Go1 hardware
- Lightweight end-to-end policy (<500 kB) operates at 50 Hz on low-cost onboard computers
Why it matters
Provides a scalable, dataset-free framework for legged robots to acquire adaptable, multi-task behaviors essential for real-world deployment in unstructured environments.
Abstract
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 lim- its 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, relax- ing 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 be- tween 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.