Uncertainty-Aware Traversability Learning and Risk-Aware Navigation for Reliable Off-Road Autonomy
This work proposes a unified framework, EVORA, to learn uncertainty-aware traction models and plan risk-aware trajectories for fast and reliable off-road navigation. The proposed approach efficiently quantifies both aleatoric and epistemic uncertainty, and leverages the learned uncertainty-aware traction model to enable risk-aware planning that handles the risk of immobilization due to uncertain terrain.