Use of neural networks and predictive control to determine the steering control behavior required for racecar simulations.

Chris Rijnders (TU Delft and Reynard Racing Cars)

The ever-increasing cost of running racecars drives the need for an accurate simulation. It can be used as a development tool (for example, to check the effect of a new wing using windtunnel data before the car has even run) or as a setup tool. (For gear ratios, spring rates, ride heights, front and rear wing settings, etc.) In this way many things can be known about the setup of the car before a race weekend starts, giving the team using it an advantage.

One of the most important components of the simulation is the driver model. The requirement is not only that it will accurately mimic human behaviour, but this `driver' must also be sensitive to changes to the car, thereby allowing the user of the simulation to see what effect any changes will have. The creation of a driver model for a simulation is complicated by the highly non-linear behavior of a racecar.

In this work, two different ways are implemented for achieving this aim: the first uses elementary neural networks, the other uses predictive control.

Chris Rijnders is a student at the Control & Simulation group of the Aerospace Engineering faculty of the Delft University of Technology in Holland. He is currently completing his masters degree with his work on the new simulation of Reynard Racing Cars.

Back to Control Seminars Page.