Automated vehicle control beyond the stability limits
- Jonathan Yan Ming Goh.
- [Stanford, California] : [Stanford University], 2019.
- Copyright notice
- Physical description
- 1 online resource.
Also available at
- Automated vehicles can access a much wider range of maneuvers - and therefore, avoid accidents in a larger number of scenarios - by operating safely beyond the stability limits. This thesis presents a series of contributions towards this goal, supported throughout by fully autonomous experiments on MARTY, a heavily-modified 1981 DMC DeLorean. First, we develop a physically insightful controller structure for automated drifting along a general path. In stark contrast with conventional driving scenarios, the speed of the vehicle is not explicitly tracked. Instead, we show that the coupling between lateral and longitudinal dynamics, often regarded as a formidable challenge for human drivers, allows velocity to be regarded as a stable zero dynamic under the imposed control law. This approach is experimentally demonstrated on a slowly-changing trajectory. The thesis then builds upon this result to present the fully autonomous execution of highly dynamic drifting maneuvers. The planning of a 'Figure 8' trajectory is formulated as a nonlinear optimization problem. We show that the key challenge is the loss of independent control over the velocity vector and vehicle body rotation rates that occurs when operating at the limits of the achievable state derivatives. Experiments confirm the effectiveness of modifications to the controller that account for this effect. To confidently venture past the well-known stability limits and utilize these transient maneuvers, however, it is critical to know the bounds on what is achievable. This thesis proposes a novel and intuitive approach for defining such an envelope, namely that the relevant criteria is avoiding the loss of steering authority over the rotational dynamics. The utility of this approach is demonstrated through a pair of experiments with and without a 'supervisor' that modifies the underlying controller. While the vehicle successfully completes the maneuver with the supervisor active, it spins out otherwise. Finally, the effectiveness of this combined approach is showcased though a series of tests on a long sustained drifting course of almost 1 kilometer in length, that includes several highly dynamic transitions, and reaches 40 degrees of sideslip, 2.5 rad/s of yaw rate, and speeds of 50km/h.
- Publication date
- Copyright date
- Submitted to the Department of Mechanical Engineering.
- Thesis Ph.D. Stanford University 2019.