Massachusetts Researchers Built AI-Powered Stunt Bicycle

A new research-stage robotic vehicle demonstrates advanced jumping and balancing capabilities in controlled testing.

Updated on Oct. 14, 2026 in Robotics

Massachusetts Researchers Built AI-Powered Stunt Bicycle

Live Poll

Would you trust an autonomous robot to deliver goods to your home?

Massachusetts researchers have developed the Ultra Mobility Vehicle, an AI-powered robot that balances on two wheels and performs complex stunts. The research-stage robot successfully executes jumps, wheelies, and front flips in laboratory environments.

Why it matters

By optimizing for agility and balance through reinforcement learning, this project explores new methods for navigating challenging terrain. These capabilities could eventually inform the design of hardware for urban delivery systems or reconnaissance tasks.

The 23.5kg robot utilizes a movable head to dynamically shift its center of mass, allowing it to perform a front flip in under 0.5 seconds. It cleared a one-metre-high platform during testing, significantly exceeding its own physical height.

The players

Ultra Mobility Vehicle

An AI-driven robotic bicycle capable of autonomous stunts and obstacle navigation.

The details

Engineers trained the Ultra Mobility Vehicle using reinforcement learning — a machine learning technique where an agent learns to make decisions by receiving rewards for successful actions in a simulated environment. The design, which draws inspiration from the movement of mountain bikers, utilizes a single wheel for balance while adjusting its frame to execute jumps. Simulations were used to iterate through dozens of structural configurations before physical validation.

October 14, 2026: The team published their design findings in Science Robotics.

The Tech Race

This development follows a pattern established by the MIT Cheetah robot project, which prioritizes dynamic agility for navigating non-flat terrain. The work marks a move toward applying high-performance movement algorithms to smaller, single-wheel-balanced architectures.

The technology currently exists as a research-stage prototype and is not available for public use or purchase. Its success in simulated and controlled environments will influence future designs for automated urban delivery and specialized reconnaissance hardware.

The takeaway

The project demonstrates that reinforcement learning can effectively control complex, non-linear mechanical systems for high-impact movement. Observers should track subsequent peer-reviewed findings regarding the vehicle's energy consumption and load-bearing performance in outdoor scenarios.