In robotics, traversability, the ability of a robot to navigate challenging terrains, remains a significant challenge due to limitations in design, perception, and computation. The hybrid wheel-legged design, exemplified by the SherpaTT rover, offers enhanced mobility in unstructured terrain. Current systems rely on sparse LiDAR-based offline traversability maps for static environments, making them unsuitable for dynamic exploration.
This research proposes an online traversability estimation pipeline using robot-centric elevation mapping for the SherpaTT rover. By leveraging the RGB-D data from a stereo camera, the method provides real-time perceptual insights and traversability assessments based on semantic and geometric features. This approach enables dynamic navigational decisions, significantly enhancing the rover's exploration capabilities.
