Discrete-Time Hybrid Automata Learning: Legged Locomotion Meets Skateboarding
Hang Liu, Sangli Teng, Ben Liu, Wei Zhang, Maani Ghaffari
2025-03-05
Summary
This paper talks about a new way to teach robots complex movements that involve both smooth motions and sudden changes, like a four-legged robot learning to skateboard
What's the problem?
Current methods for teaching robots these kinds of movements either rely on pre-programmed steps or don't understand when to switch between different types of motion. It's especially hard to teach robots complex movements without breaking them down into smaller parts first
What's the solution?
The researchers created a system called DHAL that uses artificial intelligence to learn when to switch between different types of motion. They tested it by teaching a four-legged robot to skateboard, which involves both rolling smoothly and pushing with its legs. DHAL uses special math to understand both the smooth and sudden changes in movement
Why it matters?
This matters because it could help robots learn more complex and natural movements on their own. This could lead to robots that can adapt to new situations more easily, like helping in search and rescue operations or working in factories where they need to handle different tasks
Abstract
This paper introduces Discrete-time Hybrid Automata Learning (DHAL), a framework using on-policy Reinforcement Learning to identify and execute mode-switching without trajectory segmentation or event function learning. Hybrid dynamical systems, which include continuous flow and discrete mode switching, can model robotics tasks like legged robot locomotion. Model-based methods usually depend on predefined gaits, while model-free approaches lack explicit mode-switching knowledge. Current methods identify discrete modes via segmentation before regressing continuous flow, but learning high-dimensional complex rigid body dynamics without trajectory labels or segmentation is a challenging open problem. Our approach incorporates a beta policy distribution and a multi-critic architecture to model contact-guided motions, exemplified by a challenging quadrupedal robot skateboard task. We validate our method through simulations and real-world tests, demonstrating robust performance in hybrid dynamical systems.