Yamaha Motor has introduced MOTOROiD:Λ, the third-generation concept in its ongoing exploration of advanced human-machine interaction for two-wheeled vehicles, at the Japan Mobility Show 2025 in Tokyo this month.

The MOTOROiD project began in 2017 with a proof-of-concept motorcycle that could recognize its owner, approach autonomously, and stand upright without the need for a kickstand. The 2023 MOTOROiD2 took a significant step forward by enabling real-time mutual response. The machine interpreted the rider’s weight shifts and intentions, while the rider felt the motorcycle’s subtle feedback. This created a partnership-like riding experience.

The new MOTOROiD:Λ is the first version to incorporate reinforcement learning (RL) for motion control. In simple terms, the motorcycle repeatedly practices riding maneuvers inside a virtual environment. Each time it improves balance, lean response, standing stability, or smooth parking, the algorithm is rewarded and refines the behavior. After extensive virtual training, the learned skills are transferred to the physical bike through Sim2Real technology, allowing the motorcycle to continue improving in the real world without manual reprogramming.

Reinforcement learning teaches the motorcycle to achieve perfect balance, smoother cornering, and even park itself without the rider touching the bars.

The MOTOROiD:Λ has a computer informed by an array of IMUs, so it is fully aware of lean angle, yaw, pitch, and acceleration. The IMUs are bolstered by cameras and LiDAR (Light Detection and Ranging). That allows MOTOROiD:Λ to see the ground and obstacles. With that information, the computer instructs actuators to move the steering, shift weight, and control a small balance motor.

The process begins with computer simulations. Engineers create a digital twin that perfectly replicates the motorcycle and the real world. Inside the sim, the bike attempts to perform various actions— standing still, leaning into a corner, or rolling forward slowly are examples—thousands of times per minute.

If it gets closer to the goal, the system is given a positive numerical reward. If it falls over or wobbles, it is given a negative numerical reward.

No human is writing instructions. The bike figures it out by trial and error. An example of this is when AlphaGo learned to beat humans at Go.

After millions of simulated crashes, the AI has learned a policy that uses sensor-like data to produce smooth, stable movement. The result, according to Yamaha engineers, is motion that appears more fluid and lifelike than traditional programmed robotics.

The trained neural network is then loaded into the physical motorcycle. Because the simulation was extraordinarily accurate, taking into account inputs such as tire grip, wind, and road texture, the bike can now seamlessly perform those same maneuvers on real pavement. It continues to learn in the real world, though more carefully and at a much slower pace, to reduce damaging crashes.

Yamaha designed the chassis of MOTOROiD:Λ specifically for this learning process. It has a lightweight yet durable carbon-fiber exoskeleton that protects critical components from the repeated low-speed falls that occur as AI experiments with new movements.

Reinforcement learning turns the motorcycle from a machine that simply obeys external input into one that teaches itself how to be a better partner and provide valuable assistance. The current version of MOTOROiD:Λ balances at stops without the rider putting a foot down — even on side slopes — rolls itself at walking pace in parking lots, makes lean-in feel more natural by subtly assisting instead of resisting, and continues to refine its behavior the longer it is ridden as it adapts to its owner’s style.

According to Yamaha, giving MOTOROiD:Λ a degree of independent decision-making marks a fundamental shift in rider-machine relationships. Instead of a tool that simply follows commands, MOTOROiD:Λ grows and adapts alongside its owner over time.