Cooperative Path Planning for Adversarial Target based on Neural Network and Artificial Potential Field

Abstract

Cooperative path planning for adversarial target is a challenging problem in competitive application scenarios such as military combats and sport games. In these scenarios, the goal and obstacle are dynamically moving to react against the path planning decisions just made, which makes conventional path planning methods not efficiency as before. The cooperative path planning of multi-robots is also a difficulty in this case. In this paper, a cooperative path planning method for adversarial target is proposed. The neural network is introduced to adaptively adjust the potential gain coefficients of the artificial potential field depends on the relative situation of adversarial target and cooperative partner. The neural network model is trained offline by approximately optimal samples. The method is evaluated in the 2 on 1 beyond-visual-range air combat path planning, which shows that our method significantly improves the win ratio by more than 30% comparing to the classic method.