Learning from fuzzy system demonstration: Autonomous navigation of mobile robot in static indoor environment using multimodal deep learning

Recent advances in deep learning and high-performance computational hardware, have shown great potential for solving mobile robot navigation and obstacle avoidance problems. In autonomous navigation, mobile robots use onboard sensors to perceive the environment and extract useful information to safely navigate to the target goal. Thus, fusion of multiple modalities improves the navigation system performance significantly. Multimodal deep learning networks are concerned with learning richer representations from multiple modalities which improves network's predictive ability compared to unimodal networks. In this paper, we propose a multimodal deep learning network for autonomous navigation of mobile robots in a static indoor environment. Further, a fuzzy-based behavior navigation system is designed to collect datasets for training the designed multimodal network. The evaluation of the trained models showed that the network was able to control mobile robot motion and navigate safely to the target goal. Further, the network mimics the cognitive decision-making ability of the fuzzy system and generalizes to unseen scenarios.