LiDAR-Only Navigation of UGVs in Dynamic Environments via Graph Attention Networks and Deep Reinforcement Learning

A novel deep reinforcement learning (DRL) framework is proposed for autonomous navigation of unmanned ground vehicles (UGVs) in dynamic environments using LiDAR data. Utilizing a Graph Attention Network (GAT), the framework processes high-dimensional LiDAR data into a meaningful state representation, enabling informed navigation decisions and adaptability to changing obstacles. The approach is validated through simulations of a Husky A200 UGV, demonstrating significant performance improvements over baseline TD3 algorithms. Key results include an improved success rate in trained environments and enhanced generalization in unseen scenarios, with reduced collisions and shorter navigation times. The findings highlight the potential of GAT-enhanced DRL for efficient and cost-effective autonomous navigation in real-world applications.