Intelligent Control of Swarm Robotics Employing Biomimetic Deep Learning

The collective motion of biological species has robust and flexible characteristics. Since the individual group interacts with other neighbors asymmetrically, which means pairwise interaction presents asymmetrical characteristics during motion, building model is still full challenges. Based on deep learning (DL) technology, experimental data Hemigrammus rhodostomus fish are analyzed to build an multi-parameter input. First, a Deep Neural Network (DNN) structure for designed. Then, obtained by DNN proper training. We propose novel key neighbor selection strategy, called Largest Visual Pressure Selection (LVPS) method, deal multi-neighbor interaction. information identified LVPS, uses properly trained Compared strategies, statistical properties simulated our proposed more consistent those experiments. simulation shows that method can extend large-scale aggregation control. Thereby, take advantage quite limited local collaboratively achieve motion. Finally, we demonstrate swarm robotics in platform. control simple use, applicable different scales, fast calculation. Thus, it broad application prospects fields multi-robotics control, intelligent transportation systems, saturated cluster attacks, multi-agent logistics, among fields.