Optimization based trajectory planning for autonomous racing

Sammanfattning: Autonomous driving is one of the three new technologies that are disrupting the classical vehicle industry together with electrification and connectivity. All three are pieces in the puzzle to drastically reduce the number of fatalities and injuries from traffic accidents but also to reduce the total amount of cars, reduce the polluting greenhouse gases, reduce noise pollution and completely eliminate unwanted driving. For example would most people rather rest, read or do anything else instead of driving in congested traffic. It is not small steps to take and it will have to be done incrementally as many other things. Within the vehicle industry racing has always been the natural place to push the boundaries of what is possible. Here new technologies can be tested under controlled circumstances in order to faster find the best solution to a problem.Autonomous driving is no exception, the international student competition ”Formula Student” has introduced a driverless racing class and Formula E are slowly implementing Robo Race. The fact that race cars aim to drive at the limits of what is possible enable engineers to develop algorithms that can handle these conditions even in the every day life. Because even though the situations when normal passenger cars need to perform at the limits are rare, it is at these times it can save peoples lives. When an unforeseen event occurs and a fast manoeuvre has to be done in order to avoid the accident, that is when the normal car is driving at the limits. But the other thing to take into considerations when taking new technology into the consumer market is that the cars cannot cost as much as a race car. This means simpler computers has to be used and this in turn puts a constraint on the algorithms in the car. They can not be too computationally heavy.In this thesis a controller is designed to drive as fast as possible around the track. But in contrast to existing research it is not about how much the limit of speed can be pushed but of how simple a controller can be. The controller was designed with a Model Predictive Controller (MPC) that is based on a point mass model, that resembles the Center of Gravity (CoG) of the car. A g-g diagram that describes the limits of the modeled car is used as the constraints and the cost function is to maximize the distance progressed along the track in a fix time step. Together with constraints on the track boundaries an optimization problem is giving the best possible trajectory with respect to the derived model. This trajectory is then sent to a low level controller, based on a Pure Pursuit and P controller, that is following the predicted race trajectory. Everything is done online such that implementation is possible. This controller is then compared and evaluated to a similar successful controller from the literature but which has a more complicated model and MPC formulation. The comparison is made and some notable differences are that the point mass model is behaving similar to the more complex model from the literature. Though is the hypothesis not correct since the benefits of the simplification of the model, from bicycle to point mass model, is replaced when more complex constraints has to be set up, resulting in similar performance even in computational times.A combination of the two models would probably yield the best result with acceptable computational times, this is left as future work to research.