Abstract: Each year there are approximately 40,000 fatalities on US roadways, 40% of which result from a collision with a fixed obstacle in the environment. Thousands of lives, therefore, could be saved by simply helping the driver keep the vehicle in the lane. This talk describes an approach to driver assistance based on artificial potential fields that define the lane boundaries as hazards with the minimum hazard in the center of the lane. Analogous to a marble rolling in a valley, the lanekeeping assistance system attempts to nudge the vehicle back to the minimum hazard. When the driver is tracking the lane, the car feels exactly how it would without any assistance; as the driver deviates from the center, the car gently adds an additional steering command, producing an effect much like being attached to the road with a light spring.
Mathematically, the system can be formulated in terms of Lagrangian dynamics with the lanekeeping system adding a force on top of the existing vehicle dynamics. Forming a Lyapunov function for the system, lanekeeping performance can be guaranteed even in the presence of curves or other disturbances. This formulation also gives insight into how the lanekeeping system performs at the handling limits, enabling stability and performance guarantees even when the tire friction limits are exceeded.
The mathematical results suggest that very simple algorithms can lead to highly robust vehicle control. To further support this claim, the talk presents the results of experiments involving a 1997 Chevrolet Corvette and an all-electric student-built steer-by-wire vehicle, P1. Data and video from these experiments show that the theoretical performance guarantees hold even in real-world situations when the car is pushed beyond the friction limits.