KiRAS: Keyframe Guided Self-Imitation for Robust and Adaptive Skill Learning in Quadruped Robots

Xiaoyi Wei1             Peng Zhai*,1             Jiaxin Tu1             Yueqi Zhang1             Yuqi Li1             Zonghao Zhang1             Hu Zhou2             Lihua Zhang*,1            

1
2026 IEEE International Conference on Robotics and Automation (ICRA)
multi-skill

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 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.

Skill Imitation and Transition

Skill Imitation based on Keyframes

Flexible Skill Transition on Solo-8

Solo-8 Locomotion Ability

Crossing Rough Terrain

Using Different Skills to Traverse Different Terrain

Skill Switching Test

Unitree Go1 Biped Skill Test


Our Previous Work: PASIST

Continuous Control of Diverse Skills in Quadruped Robots Without Complete Expert Datasets

Jiaxin Tu       Xiaoyi Wei       Yueqi Zhang       Taixian Hou       Xiaofei Gao       Zhiyan Dong       Peng Zhai       Lihua Zhang
2025 IEEE International Conference on Robotics and Automation (ICRA)
  • This paper applies Generative Adversarial Self-Imitation Learning (GASIL) to quadruped robot skill learning for the first time, and designs a new metric by integrating task rewards with Dynamic Time Warping (DTW) values to realize autonomous selection and extraction of high-quality trajectories without complete expert datasets.
  • It serves as the theoretical foundation of KiRAS.

BibTeX


      @inproceedings{tu2025continuous,
        title={Continuous Control of Diverse Skills in Quadruped Robots Without Complete Expert Datasets},
        author={Tu, Jiaxin and Wei, Xiaoyi and Zhang, Yueqi and Hou, Taixian and Gao, Xiaofei and Dong, Zhiyan and Zhai, Peng and Zhang, Lihua},
        booktitle={2025 IEEE International Conference on Robotics and Automation (ICRA)},
        pages={11191--11197},
        year={2025},
        organization={IEEE}
      }