Remote Vehicle Tracking with Location-dependent Communication Cost

초록

Location-dependent communication cost poses unique challenges in tracking vehicle due to the time-varying transmission cost according to the vehicle's movement. In this work, we show that the traditional single-threshold policy is not suitable for vehicle tracking with location-dependent cost, and investigate the characteristics of optimal policy through two interesting heuristics: multi-threshold policy that employs multiple thresholds for different communication costs, and Deep Q-Network (DQN) policy that exploits Reinforcement Learning techniques. Through extensive simulations, we demonstrate that they exhibit different decision-making strategies. In particular, the multi-threshold policy offers faster convergence and the DQN policy achieves better long-term performance. Our results highlight that it is of great significance to adjust vehicle's transmission policy in response to the varying communication costs. © 1967-2012 IEEE.

키워드