Skateboarding Robot Flips Boards and Minds

Researchers at the University of Michigan’s Computational Autonomy and Robotics Laboratory (CURLY Lab), alongside China’s Southern University of Science and Technology (SUSTech), have developed an artificial intelligence (AI) framework that enables legged robots to skateboard.

The system, called Discrete-Time Hybrid Automata Learning (DHAL), allows robots to handle complex, contact-heavy tasks like balancing, pushing off and gliding on a moving skateboard.

Unlike previous methods, DHAL doesn’t require manual labelling of movement transitions. Instead, it uses reinforcement learning to autonomously identify and respond to changes in dynamics – whether it’s shifting weight to push forward or stabilising during a glide.

These are known as “hybrid dynamics”, which combine smooth, continuous motion with abrupt and discrete changes, much like a bouncing ball hitting the ground.

Sangli Teng, co-author of the study, explained: “Existing quadrupedal locomotion approaches do not consider contact-rich interaction with objectives, such as skateboarding. Our work was aimed at designing a pipeline for such contact-guided tasks that are worth studying.”

To test the AI system, the CURLY Lab and SUSTech team used a Unitree Go1 robot, which learned to mount a skateboard, ride it across various surfaces and even pull a small cart – all without joystick controls. In trials, the automaton reliably switched between pushing, gliding and balancing phases with minimal errors, outperforming traditional algorithms.

Beyond skateboarding, DHAL could have broader applications, such as warehouse automation and robotic delivery systems. The system could easily be applied in environments where terrain and tasks require dynamic physical interaction.

The team’s next goal: applying DHAL to dexterous tasks such as manipulating objects with robotic fingers.

Image Credit: Source