Beyond the Limits: Controller Design and Trajectory Generation for Autonomous Drifting Cars
MARTY is the world's first - and only - autonomous electric drifting DeLorean. It is the physical manifestation of my PhD thesis at Stanford University's Dynamic Design Lab, led by Prof. Chris Gerdes.
The video above shows a recent version of the autonomous controller + trajectory planner in action. In order to deal with the difficult dynamics of drifting, the controller draws on principles from nonlinear control. To improve robustness and performance, it also incorporates elements from adaptive control, namely online parameter estimation. It follows a highly dynamic trajectory planned offline as a solution to a constrained nonlinear optimization problem. Underpining all of this is an appropriate choice of vehicle dynamics model that balances fidelity, simplicity, and intuition.
In this example, the trajectory transitions from drifting left at -40 degrees of sideslip to drifting right at +40 degrees of sideslip in about a second. You can see the steering swing from lock to lock in about the same amount of time. Throughout the trajectory, we track the path to within +/- 40cm, and sideslip to +/- 5 degrees.