Real-time Trajectory Generation and Control for Autonomous Vehicles

Abstract

For many years robotics has provided solutions to problems which are either dull, dirty, or dangerous. Advancements in computer processing, sensors, and actuator technology made it possible for some of these robotic systems to require less and less human interaction. However, autonomy still poses many unsolved challenges. This dissertation presents a set of algorithms to control high-performance autonomous vehicles in real-time. All algorithms are designed with actual application in mind. They have been implemented and tested successfully on real land-based or airborne autonomous vehicles.

The first chapter describes an algorithm to calculate near-optimal minimum time trajectories for four wheeled omnidirectional vehicles, which can be used as part of a high-level path planner. The algorithm is based on a relaxed optimal control problem. It takes limited friction and vehicle dynamics into account, as encountered in high-performance omnidirectional vehicles. The low computational complexity makes the application in real-time feasible. An implementation of the algorithm on a real vehicle is presented and discussed.

The second chapter presents a cooperative decentralized path-planning algorithm for a group of autonomous agents that provides guaranteed collision-free trajectories in real-time. The algorithm is robust with respect to arbitrary delays in the wireless traffic, possible sources being transmission time and error correction. Agents move on reserved areas which are guaranteed not to intersect, therefore ensuring safety. A handshaking procedure guarantees recent information states for the agents. Conflicts between agents are resolved by a cost-based negotiation process. The basic algorithm is augmented by the introduction of waypoints, which increase performance at the cost of additional wireless traffic. An implementation of the algorithm is tested in simulation and successfully applied to a real system of autonomous robots. The results are presented and discussed.

The third chapter presents an algorithm to iteratively perform an aggressive maneuver, i.e. drive a system quickly from one state to another. A simple model which captures the essential features of the system is used to compute the reference trajectory as the solution of an optimal control problem. Based on a lifted domain description of that same model an iterative learning controller is synthesized by solving a linear least-squares problem. The controller adjusts a feedforward signal using the results of experiments with the system. The non-causality of the approach makes it possible to anticipate recurring disturbances. Computational requirements are modest, allowing controller update in real-time. The experience gained from successful maneuvers can be used to adjust the model, which significantly reduces transients when performing similar motions. The algorithm is successfully applied to a real quadrotor unmanned aerial vehicle. The results are presented and discussed.