At the Tokyo Motor Show last year, Yamaha introduced its Motobot, a humanoid robot that rides a YZF-R1. Video was released showing the Motobot riding in an open area, mostly going in a straight line, but the intention is for Motobot to ride the bike on track sometime next year and eventually beat the lap times of none other than Valentino Rossi. That is a tall order.
Autonomous, or self-driving cars have made steady progress over the last several years. Audi, in conjunction with Stanford University's Dynamic Design Lab, has been conducting a program with a self-driving TTS dubbed Shelley and testing at Thunderhill Raceway Park in California. Currently, the car is as fast as a human driver - and not just any human driver, but David Vodden, CEO of the track and an Amateur touring-class champion. The car is programmed to drive the "ideal" GPS line around the racetrack, using sensors and feedback to maximize speed while retaining control.
Shelley is definitely impressive, but bikes are very different from cars, as we know. On top of the basic inputs of steering, throttle and brake that cars and bikes share, a motorcycle rider can significantly influence performance with his body position and movements. If Motobot is going to be as fast as Rossi around the track, it makes sense that the robot would need the ability to move around on the bike like Rossi.
It will be interesting to see how this is incorporated into the Motobot, or even if Yamaha does give it movement capabilities. The trouble here is, just what do you tell it to do? Audi and automobile companies can rely on a complete slate of driver-input data to help program their autonomous vehicles, and use that data to help program the car. In a press release issued by Stanford, Mechanical Engineering Associate Professor Chris Gerdes said, "We need to know what the best drivers do that makes them so successful. If we can pair that with the vehicle dynamics data, we can better use the car's capabilities." Recording an analyzing a driver's inputs is relatively straightforward, but for a motorcycle rider it's much more complex.
Some inputs and movements are obvious on the motorcycle, such as countersteering, hanging off while turning or moving rearward under braking and forward under acceleration. But it's the subtleties that
can add up to big chunks of time at the track, and those subtleties are almost impossible to measure. Ask 10 riders how they use their body to steer the motorcycle, or what they are trying to accomplish using the "leg dangle," and you will get 10 different answers. Programming a robot to perform those movements, let alone physically react well to feedback from sensors such as an IMU, would be a mountainous task.
According to reports from Thunderhill, human drivers still have an advantage over Shelley in that they can adapt to a wider variety of situations, whereas Shelley only knows to keep to its ideal line. So, for example, if Shelley overshoots a corner, it attempts to return to its programmed course as soon as possible, perhaps losing a fair amount of time. A human driver, however, may reduce that loss by changing the next turn's entry - and even learn a superior line from the experience.
While Motobot may not succeed in beating Rossi's times at the track for whatever reason, Yamaha will still learn plenty from the project; according to the company, the technology learned will be applied to more advanced safety systems in the future. Hopefully Yamaha will release more details on the project as it proceeds. For me, it's the relationship between the rider's (or robot's) inputs and the bike's dynamics that is the most interesting. Many aspects of chassis setup, electronic controls and riding skills rely on optimizing that interface between rider and bike, and that is the key to making improvements in those areas, whether you are trying to set a fast lap at the track or be safe on a country road.