Technology for Dynamic Path Planning of Autonomous Mobile Robots Based on Deep Learning and Multi-Sensor Fusion

Because autonomous mobile robots operating in complex dynamic environments encounter several difficult problems, namely sudden dynamic obstacles, heterogeneous information from multiple sensors, and environmental uncertainty, this paper presents a thorough investigation of using deep learning and multi-sensor fusion for dynamic path planning, and accordingly designs a multi-dimensional perception layer based on visual cameras, lidar, infrared detectors, and IMUs. The method processes data by means of hybrid filtering and temporal-spatial synchronization, then uses CNN, LSTM, and Transformer networks to extract cross-modal features and perform dynamic feature fusion, while the core of the approach is the application of deep reinforcement learning models DQN and PPO to build a natural, tight-coupled closed-loop system for perception, decision-making, and execution, hence achieving real-time optimal obstacle avoidance and path planning. More importantly, the method has excellent adaptability to dynamic environments and strong robustness against interference.