Adaptive Dynamic Programming in a Subspace for Some Discrete-Time Vehicle Systems

Vehicles of all sorts are important tools of human activities. Modern autonomous vehicles are built to perform complicated tasks that are beyond their basic function as transportation tools. Examples include robots that autonomously explore the unknown environments, and self-driving vehicles that safely navigate on the highways. Because their working environments are highly dynamic or unknown, autonomous vehicles need to make decisions and react to their changing environments at every time instance. This decision-making problem can be solved using constrained optimal control methods. However, solving such problems in real time is prohibitive on most vehicles because current methods take a large amount of computational resources and most vehicles lack that level of computational power. In this dissertation, a new adaptive dynamic programming method with reduced computation requirement is developed to solve this type of problems. Based on a bioinspired search strategy and the knowledge of vehicle dynamics, the new method can help vehicles make decisions in real time with a fraction of the computational resources required by other typical constrained optimal control methods. An unmanned aerial vehicle flight control problem and a ground vehicle obstacle avoidance problem are used to test the performance of the new method in simulation. A scouting robot has successfully adopted this new method for its navigation in a local farm