Co-Optimization of Motion, Communication, and Sensing in Real Wireless Channel Environments via Monte Carlo Tree Search

• Published in: IEEE transactions on control of network systems 2022-09, Vol.9 (3), p.1493-1505
• Main Authors: Cai, Hong, Mostofi, Yasamin
• Format: Article
• Language: English
• Subjects: co-optimiza- tion of motion; Communication; Communication and sensing; Convergence; Data collection; Data transmission; Decisions; Energy costs; Fading; Fading channels; Markov processes; Monte Carlo simulation; Optimization; Robot dynamics; Robot sensing systems; robotics; Robots; sensor networks; Sensors; Task analysis; Wireless communication; Wireless communications; Wireless sensor networks
• ISSN: 2325-5870; 2325-5870; 2372-2533
• DOI: 10.1109/TCNS.2022.3158746

We consider the problem where a robot navigates from a start position to a destination and needs to sense some sites along the way. The robot collects data when sensing each site and needs to transmit all collected data to a remote station by the end of its trip, as it moves along its path, under time/energy constraints, and while operating in real wireless fading environments. Our goal is to minimize the robot's total motion and communication energy costs by co-optimizing its path, data transmission along the path, and sensing decisions, under given constraints and while considering the stochastic wireless channel. In this paper, we show how to solve this co-optimization problem efficiently and with performance guarantees. More specifically, we formulate a specially-designed Markov Decision Process (MDP) and utilize Monte Carlo Tree Search (MCTS) to efficiently and optimally solve it. While the co-dependence of communication, sensing, and motion decisions makes this joint optimization challenging, we show that by considering the transmission optimization in the terminal reward and motion actions in the state transitions, we can iteratively optimize the sensing/motion and the communication parts in different stages of MCTS, in a way that allows us to equivalently solve the original co-optimization problem efficiently. We mathematically prove the convergence of our approach, characterize its convergence speed, and derive key properties of the optimum solution. We extensively evaluate our approach in realistic wireless environments where the channel experiences path loss, shadowing, and multi-path fading and is unknown to the robot.