eduzhai > Applied Sciences > Engineering >

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

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 15 pages

Abstract: This paper presents a novel model-reference reinforcement learning algorithmfor the intelligent tracking control of uncertain autonomous surface vehicleswith collision avoidance. The proposed control algorithm combines aconventional control method with reinforcement learning to enhance controlaccuracy and intelligence. In the proposed control design, a nominal system isconsidered for the design of a baseline tracking controller using aconventional control approach. The nominal system also defines the desiredbehaviour of uncertain autonomous surface vehicles in an obstacle-freeenvironment. Thanks to reinforcement learning, the overall tracking controlleris capable of compensating for model uncertainties and achieving collisionavoidance at the same time in environments with obstacles. In comparison totraditional deep reinforcement learning methods, our proposed learning-basedcontrol can provide stability guarantees and better sample efficiency. Wedemonstrate the performance of the new algorithm using an example of autonomoussurface vehicles.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×