DQN based coverage control for multi‐agent system in line intersection region

Abstract

AbstractGenerally, the coverage control is studied in a convex region, in which the agent kinematics and the coverage environment both have strong limitations. It is difficult to directly apply these results to practical scenarios, such as the road environment or indoor environment. In this study, the multi‐agent coverage control problems in a line intersection region is investigated, where the agents can only move along the given lines. To present the agents motion in this line intersection region, the moving directions and velocities of the agents are analyzed in the first part. Then, the coverage control model for the multi‐agent system in line intersection region is presented, in which the cost function is provided based on the agent's minimum moving distance and the agent motions are used as the constraints. To solve this constrained coverage problem, the deep Q‐learning network (DQN) is employed to find the optimal positions for each agent in the line intersection region. In final, numerical simulations are presented to validate the feasibility and effectiveness of proposed approaches.