Neural Dynamic Collision-Avoidance Strategy for Robots Based on Evaluation of Threat Degree

For the autonomous collision-avoidance of mobile robots in dynamic environments, a neural collision-avoidance decision system consisting of a sensory module, a threat-avoidance module and a target-directed module is constructed. The relative velocity between threats and robot, effectual detection range of lidar, and prediction azimuth of threat within the sampling period are investigated, and their effects on the maneuvering efficiency and security of collision avoidance are analyzed deeply. Based on this, the risk level of each maneuvering area is determined, and a quantitative evaluation model of threat degree is established. Then, it is introduced into the local decision maker to correct the output of each node. Furthermore, a neural dynamic collision-avoidance method based on evaluation of threat degree of the maneuvering strategy is proposed. Simulation experiment results indicate that, compared with traditional neural network based navigation approaches, the proposed approach can optimize avoidance paths with a lower cost.