Legged robots skate successfully using new learning framework

Researchers at the University of Michigan and Southern University of Science and Technology have developed a new framework that helps legged robots successfully skateboard. This framework is based on reinforcement learning, which allows robots to perform complex movements while interacting with objects. The research team created a method called discrete-time hybrid automata learning (DHAL). This new approach enables robots to handle both smooth and sudden movements, which is crucial for tasks like skateboarding. Current methods struggle with these transitions, but DHAL can identify them without needing prior knowledge about the robot's movements. In initial tests, the robots were able to step onto a skateboard and glide while pulling a small cart. This new capability could help robots deliver packages in various environments, such as urban areas and offices. The research team is considering applying the DHAL framework to other tasks, like manipulating objects with multiple arms. The scientists believe this technology could greatly improve how robots operate in real-world scenarios.