A Survey of Traversability Estimation for Mobile Robots

Traversability illustrates the difficulty of driving through a specific region and encompasses suitability terrain for traverse based on its physical properties, such as slope roughness, surface condition, etc. In this survey we highlight merits limitations all major steps in evolution traversability estimation techniques, covering both non-trainable machine-learning methods, leading up to recent proliferation deep learning literature. We discuss how nascence Deep Learning has created an opportunity radical improvement estimation. Finally, self-supervised can help satisfy methods’ increased need (challenging acquire label) large-scale datasets.