Multi-agent Collaborative Target Search Based on Curiosity Intrinsic Motivation

Multi-agent target search is an important research direction in the multi-agent field, and deep reinforcement learning has become one of the popular methods to study it. In the process of multi-agent collaborative target search using deep reinforcement learning, the agents face the sparse reward problem, which makes the learning difficult to converge. Therefore, this paper proposes a multi-agent collaborative target search method based on MADDPG with curiosity intrinsic motivation (MADDPG-C). Multi-agent curiosity intrinsic motivation module is designed and added to the MADDPG algorithm. The curiosity reward is taken as the intrinsic reward of the agent to make up for the lack of motivation of the agent to explore. A global reward function based on curiosity reward and environment reward is designed to solve the problem of sparse reward. We designed a simulation experiment to verify our algorithm, and the simulation results show that the agent can learn a good search strategy, and at the same time, the algorithm in this paper has great advantages in convergence speed and stability, as well as improving the target search efficiency.