This thesis concentrates on the collaborative decision-making and control between several intelligent agents in an intelligent inspection system. It mostly explores cooperative target detection and obstacle avoidance routing plans of multi-agent systems. Cooperative target detection method has adopted YOLOv3 as its core network and combines double-threshold decision-making with D-S evidence theory to improve detection probability. Obstacle avoidance planning method alters the artificial potential field method so that the inspection robot is able to independently perform obstacle avoidance. Results of experiments prove that the mAP of the cooperative target detection algorithm is higher than the original algorithm by 0.6-10.2, or 0.733, indicating that it is possible to accurately identify various types of targets. The mean length of the routes produced by the obstacle-avoidance path planning algorithm is 24 grids, implying that the robot can navigate and evade obstacles efficiently. The model suggested in this paper allows recognizing the object to avoid and the planning of the obstacle-avoidance path, thus promoting collaborative decision-making and coordinated control of several intelligent agents in smart inspection situations.
