Heightmap

DreamFlow is a deep RL framework that uses conditional flow matching to extend a robot’s perceptual range beyond its sensor FOV, enabling proactive navigation around unseen obstacles in cluttered environments.

This research introduces an approach to enhancing autonomous mobile robot navigation in diverse outdoor environments. We developed a method that transforms 3D LiDAR point cloud data into grayscale heightmaps, enabling more accurate assessment of ground textures. Our system classifies terrains in both static and dynamic environments, incorporating IMU data to account for robot motion influenced by terrain. The method demonstrates superior performance in texture analysis compared to direct point cloud analysis techniques, significantly improving the ability of mobile robots to navigate safely and efficiently across various outdoor terrains. This advancement is crucial for protecting sensitive equipment and cargo while expanding the operational capabilities of autonomous robots in complex, real-world settings.