PROJECT TITLE :

Model-Reference Reinforcement Learning for Collision-Free Tracking Control of Autonomous Surface Vehicles

ABSTRACT:

In this paper, a novel model-reference reinforcement learning algorithm for intelligent tracking control of uncertain autonomous surface vehicles with collision avoidance is presented. The algorithm was developed by the authors of this paper. In order to improve both the accuracy and intelligence of control, the proposed control algorithm integrates reinforcement learning with a more traditional method of control. In the control design that has been proposed, a nominal system is taken into consideration for the design of a baseline tracking controller through the application of a conventional control strategy. Additionally, the nominal system specifies the behavior that should be expected from uncertain autonomous surface vehicles when operating in an environment free of obstructions. Because it uses reinforcement learning, the overall tracking controller can achieve both collision avoidance and compensation for model uncertainties simultaneously in environments that contain obstacles. This is made possible by the presence of the obstacles. In contrast to more conventional approaches to Deep Reinforcement Learning, the learning-based control approach that we have proposed can provide stability guarantees while also improving sample efficiency. Using the performance of autonomous surface vehicles as an example, we show how well the new algorithm works.