MUJICA: Multi-Skill Unified Joint Integration of Control Architecture for Wheeled-Legged Robots
Yuqi LI, Peng Zhai, Yueqi ZHang, Xiaoyi Wei, Quancheng Qian, Zhengxu He, Qianxiang Yu, Lihua ZHang
AI summary
Problem
Current control frameworks for wheeled-legged robots struggle to unify diverse, high-difficulty skills under strict real-world motor constraints, often relying on manual switching or simplified actuation models that limit safety and performance.
Approach
MUJICA employs a unified, fully proprioceptive reinforcement learning framework that jointly trains multiple locomotion skills in a single policy, incorporates hard velocity- and position-dependent DC-motor constraints for safety, and uses a learned high-level skill selector to automate transitions based on internal state estimates.
Key results
- Unified blind policy jointly learns omnidirectional locomotion, high platform climbing, and fall recovery
- Hard DC-motor constraint modeling ensures safe deployment and maximizes motor torque exploitation
- Proprioception-based skill selector automates adaptive transitions without external perception
- Validated via extensive simulation and zero-shot real-world experiments on the Unitree Go2-W robot
Why it matters
Provides a safe, adaptive, and unified control framework that unlocks the full mobility potential of wheeled-legged robots for complex real-world tasks like inspection and disaster response.
Abstract
Wheeled-legged robots hold promise for traversing complex terrains and offer superior mobility compared to legged robots. However, wheeled-legged robots must effectively balance both wheeled driving and legged control. Furthermore, due to noisy proprioceptive sensing and real-world motor constraints, realizing robust and adaptive locomotion at peak performance of motors remains challenging. We propose the Multi-skill Unified Joint Integration of Control Architecture (MUJICA), a unified, fully proprioceptive control framework for wheeled-legged robots that integrates diverse low-level skills—including omnidirectional moving, high platform climb- ing, and fall recovery—within a single policy. All skills, dis- tinguished by unique indicator variables, are trained jointly with accurate DC-motor constraint modeling. Additionally, a high-level skill selector is learned to dynamically choose the optimal skill based solely on proprioceptions, enabling adaptive responses to the surrounding environment. Therefore, MUJICA enhances sim-to-real robustness and enables seamless transi- tions across diverse locomotion modes, facilitating autonomous adjustment to the environment. We validate our framework in both simulation and real-world experiments on the Uni- tree Go2-W robot, demonstrating significant improvements in adaptability and task success in unstructured environments.