If you look at the driver-assistance systems in cars compared to what is available for motorcycles, we basically only have ABS and traction control. If the rider brakes too hard, the tires tend to lock. The system can react, but by then it is already a critical situation.
We want to bring more knowledge about the vehicle's dynamics into the motorcycle to predict the future and determine if the rider is able to maintain his speed, his roll angle, and his position on the road so we can prevent critical situations.
We built a dynamic model of the motorcycle. The self-riding BMW R1200GS proves our model is real-time capable and detailed enough that it can cope with all riding dynamics. Now we have a tool to work with.
We can implement these systems without directly interfering with the vehicle's dynamics; it can run in the background like a guardian angel. For example: You wouldn't have to switch gears within an urban scenario; or the bike could help you with low-speed stabilization.
If the rider is in a safe situation, nothing will happen. But if he is approaching a curve at too high of a speed, for example, we can warn him to enhance safety before he gets into that critical situation. This is a demonstration of our basic research.
/arc-anglerfish-arc2-prod-bonnier.s3.amazonaws.com/public/7T62KVK4CK5EIM5ADTIGBEDUDM.jpg)
Under no circumstances will there be a self-riding BMW motorcycle on the street. It is only a showcase to prove what the motorcycle could be capable of technically. We have a model that we can use for future rider-assistance systems.
This project started with my Ph.D. thesis in December of 2013. In '14, we started the project with simulations, doing all the calculations and building up the differential equations. In 2015, we started building the motorcycle. The steering actuator took a lot of time.
/arc-anglerfish-arc2-prod-bonnier.s3.amazonaws.com/public/QQGF3CR5AJCVYENX7NZPHZDMCA.jpg)
About one year later, we started with our first test rides. Then we tried different controllers for lean angle, position, and trajectory following. The highest control we have developed is the model-predictive controller. With that, we get human-like riding.
If you tell the motorcycle, "Here is the street. It is 4 meters in width. You're right here, now go." It has different objectives that it tries to optimize. For example, it tries to stay more in the middle of the road than on the outside, but it also tries to minimize lean angle in curves.
We had knowledge exchanges with the automated passenger-car guys, and they are using the same methods—the model-predictive control, for example. But you can't just copy it and put it on a motorcycle. You have to reformulate the whole problem.
It helps us a lot to have the greater brother of BMW passenger cars in the background. We could, for example, use the same sensors and hardware on the motorcycle. But we are not using the same source code.
We have started a whole new department for motorcycle safety. This project started with just me, but now there are a few more people in the workshop. If we develop something new, other departments, like chassis, might be interested as well.
