Autonomous Collision Avoidance as a Markov Decision Process
Hossein Rastgoftar, Xiangyu Ni, Ella M. Atkins · Volume 2: Mechatronics; Mechatronics and Controls in Advanced Manufacturing; Modeling and Control of Automotive Systems and Combustion Engines; Modeling and Validation; Motion and Vibration Control Applications; Multi-Agent and Networked Systems; Path Planning and Motion Control; Robot Manipulators; Sensors and Actuators; Tracking Control Systems; Uncertain Systems and Robustness; Unmanned, Ground and Surface Robotics; Vehicle Dynamic Controls; Vehicle Dynamics and Traffic Control · 2016
In this paper, we apply a Markov Decision Process (MDP) to assure collision avoidance and specify optimal paths for evolution of an agent with nonlinear dynamics. We consider evolution of an autonomous wheeled robot in a motion field where a malicious agent has stochastic transitions over the field. By knowing the history of malicious agent transition as well as initial and goal destinations of the autonomous robot, the optimal path is obtained using the Bellman equation. We propose a novel finite time controller that assures reachability of desired way-points along the optimal path, where the optimal path is obtained without deriving the Hamiltonian.