ScaTE: A Scalable Framework for Self- Supervised Traversability Estimation in Unstructured Environments

For the safe and successful navigation of autonomous vehicles in unstructured environments, traversability terrain should vary based on driving capabilities vehicles. Actual experience can be utilized a self-supervised fashion to learn vehicle-specific traversability. However, existing methods for learning are not highly scalable various In this work, we introduce framework traversability, which directly from vehicle-terrain interaction without any human supervision. We train neural network that predicts proprioceptive vehicle would undergo 3D point clouds. Using novel PU method, simultaneously identifies non-traversable regions where estimations overconfident. With data gathered simulation real world, show our is capable By integrating with model predictive controller, demonstrate estimated results effective enables distinct maneuvers characteristics addition, experimental validate ability method identify avoid regions.