TAIPEI (Taiwan News) — A team at National Taiwan Normal University created 3D digital models of a humanoid robot and a scooter, then used machine learning and AI technology to teach the robot to ride the vehicle.

Jacky Baltes, a professor in the university’s Department of Electrical Engineering and leader of the team, said two-wheeled vehicles are harder to balance than four-wheeled ones when stationary or moving at low speeds. He said the team aims to enable the robot to demonstrate balance and steering control without modifying the specifications of the electric scooter, per ETtoday.

The team designed two control methods to operate the robot and tested them at different speeds and rotation rates. Test conditions included maintaining balance while moving, recovering quickly from disturbances, and navigating winding paths.

Among the models tested, the robot trained with machine learning showed an average performance improvement of about 52%.

Baltes said the team trained the robot using Nvidia Isaac Gym, a GPU-based simulation platform. Leveraging GPU computing allowed the system to perform tasks that once required thousands of CPU cores, making motion training more efficient on a single GPU.

The team has verified the robot’s scooter-riding capability in the laboratory and plans to adapt the technology for testing in larger, real-world environments.

Baltes added that teaching robots various skills is crucial for developing general-purpose robots. Unlike industrial robots, humanoid robots are designed to integrate into human environments. With advances in AI and machine learning, these robots are becoming capable of handling increasingly complex tasks, potentially assisting in disaster relief, senior care, and daily life.