Path planning for an automated vehicle using professional racing techniques
- A large part of the appeal of automated vehicles is their potential for improving safety of the average driver, especially in emergency situations. This objective raises a variety of research challenges such as sensing, perception, and human-machine interfaces. However, many accidents occur because the car has reached the physical limits of its capability often because rain or ice lowers those limits. This problem has seen much less attention in automated driving research. One way to develop an automated vehicle that can perform safely at the limits of its capability is to mimic the behavior of racecar drivers, who routinely face this challenge. The path they follow while driving at the limits is known as the racing line. When describing the racing line, some racecar drivers use basic curve shapes which could be generalized in a path planner. Piero Taruffi, a former professional driver who won the 1952 Swiss Grand Prix, describes his approach to designing a successful racing line for various turns by connecting a pair of straights using simple geometric curves. Taruffi specifically uses straights, arcs, and ``connecting curves that smoothly transition [between arcs and straights]" to describe the path. For simplicity, clothoids are chosen to emulate the connecting curves described by Taruffi. This thesis will demonstrate that clothoids, straights and arcs can reasonably describe racing trajectories measured from a skilled driver while driving at the limits. From Taruffi's description of connecting each turn with a pair of straights, the curvature of the path is used to approximate the locations of all the straights in the racing line. Then, the shape of each turn is formed by fitting clothoids and arcs to the path curvature between each pair of straights. Finally, a gradient descent approach is used to improve the geometric fit so that the maximum fitting error is well within the lap-to-lap repeatability of the skilled driver. Clothoids, arcs, and straights are then used in a design process to generate a synthesized trajectory comparable to a professional driver. Using the centerline as an initial reference path, straights are found along the track where the curvature is relatively low. Each turn is connected with clothoid and arc segments to form a nominal path around the track. The performance of a given path can be quantified by calculating its simulated lap time using the curvature of the path and a simple friction circle model. The initial path is transformed into a racing line using a simple gradient descent method with a set of heuristics. The geometry, curvature, and simulated laps times are compared to a skilled driver's measured racing line. The synthesized path is then implemented on an automated racecar and is able to operate at the defined limits of handling. The estimate of friction is an important parameter that defines how fast or aggressively the automated racecar drives the desired path. The higher the available friction the faster a racecar can travel along the desired path. For a given turn, if the estimated friction is too high, the car will understeer or oversteer away from the desired path by a substantial distance. If the estimated friction is consistently too low around the track, the racecar is not utilizing all available force and significant lap time is sacrificed. Even if the initial estimate of friction is perfect when the automated racecar begins a race, track conditions and tire temperature constantly change, affecting the available friction. In order to maximize the performance of the car without exceeding the physical limits of the tires, a recovery trajectory is generated in real time when the vehicle has entered the turn too fast. The original path is modified by adding an arc to the turn in front of the racecar. This change results in a trajectory that allows the car to slow to a safe speed early in the turn. Experimental results demonstrate that the stability, deviation from the original path and lap time are improved by re-planning the path instead of using the original path. The work in this dissertation demonstrates that path planning and re-planning at the limits of handling can be accomplished using a simple set of curves. The approaches discussed could influence future path planning systems for automated cars to avoid danger at the limits of handling.
|Type of resource
|electronic; electronic resource; remote
|1 online resource.
|Theodosis, Paul Alan
|Stanford University, Department of Mechanical Engineering.
|Gerdes, J. Christian
|Gerdes, J. Christian
|Cutkosky, Mark R
|Cutkosky, Mark R
|Statement of responsibility
|Paul Alan Theodosis.
|Submitted to the Department of Mechanical Engineering.
|Thesis (Ph.D.)--Stanford University, 2014.
- © 2014 by Paul Alan Theodosis
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...