Learn to Navigate Maplessly With Varied LiDAR Configurations: A Support Point-Based Approach

Deep reinforcement learning (DRL) demonstrates great potential in mapless navigation domain. However, such a model is normally restricted to fixed configuration of the range sensor because its input format fixed. In this letter, we propose DRL that can address data obtained from different sensors with installation positions. Our first extracts goal-directed features each obstacle point. Subsequently, it chooses global all point-feature candidates and uses these for final decision. As only few points are used support decision, refer as our approach point-based (SPN). handle LiDAR setups good performance simulation real-world experiments. Moreover, shows crowded scenarios small obstacles when using high-resolution LiDAR.