Multi-objective Sensor Management Method Based on Twin Delayed Deep Deterministic policy gradient algorithm

To solve the problem of sensor management in multi-target tracking optimization, this paper proposes a multi-target sensor management strategy method based on Deep reinforcement learning based on stochastic finite set theory. Firstly, based on the partially observable Markov decision process (POMDP) theory, the basic method of random finite set multi-target sensor management is presented. Secondly, a sensor management strategy framework based on multi-objective state estimation optimization based on deep reinforcement learning is proposed. Finally, the information gain is used as the reward function to solve the optimal sensor management strategy. The simulation based on OSPA distance evaluation verifies that the proposed sensor management method has better estimation effect on multi-target motion state than other algorithms.