University of Tokyo Humanoid Musashi Demonstrates Driving a Car at Low Speed

The University of Tokyo’s humanoid Musashi demonstrated driving a standard car at low speed by physically operating its steering wheel, pedals and turn signals. According to a report published September 12, 2026, the paper appeared in Advanced Intelligent Systems and said the robot recognized traffic signals and steered at intersections. The report characterized this as research into whether a general-purpose humanoid can use controls designed for people, not a demonstration of operation in traffic at normal speeds.

The University of Tokyo’s demonstration involved a humanoid using existing controls designed for people, rather than fitting a vehicle with new autonomous-driving equipment. This points to possible applications that do not require vehicle modifications, but practical viability cannot be assessed without data on success rates or the number of interventions.

What is new in this report is that the humanoid reproduced driving behavior by using a car’s existing controls as they are, rather than simply standing, walking or grasping. However, this is a research demonstration of whether a robot can replace a person at an interface designed for people, not an approach that adds equipment to a vehicle to make it autonomous; it is also an extension of an established trajectory. In other words, the focus is on the fit between the robot’s body and the controls, rather than the vehicle’s control systems. With no related coverage in this newspaper, the published report points to a focus less on “whether it can drive” than on “how much of the human operating space a humanoid can take on.” If it can use existing mechanical inputs such as the steering wheel, pedals and turn signals, there may be scope for applications requiring minimal vehicle modifications. But the report provides no success rates or intervention data, so the degree of reproducibility is unclear. It also says the robot recognized signals at intersections, but gives no record of driving at normal speeds. It remains unclear which part of perception, operation or decision-making is the bottleneck. At the industry level, this case could shift the competitive focus in autonomous driving slightly from software toward body dimensions, degrees of joint freedom, contact precision with input devices, and the integration of visual recognition and control. However, the report is limited to the research stage and does not yet support discussion of mass production or commercialization. The next points to check are driving speed, success rate, number of interventions, which vehicle models were used to reproduce the results, and which actions were most difficult compared with human driving.

In Japan, vehicle controls, robot body design, and the integration of visual recognition and control form the foundation for this type of research. This report is significant in showing the connection between those areas, but the work is not yet at a stage that directly leads to commercial vehicles or public-road operation.