Augmentations for Vegetation, Obstacle, and Indentation Detection and Segmentation (AVOIDS)

Vegetation and negative obstacles (potholes) are some of the most challenging yet significant features of the environment that an off-road, self-driving robotic combat vehicle (RCV) needs to consider when navigating. Failure to correctly identify these obstacles or separate them from road surfaces and rocks may result in the immobilization of or damage to the vehicle. Contemporary self-driving navigation approaches relying on lidar as their sole source of sensory input are ill-equipped for this task as they lack the fidelity to recognize these features and use emitting projectors that can reveal the vehicleÆs presence. Academic work has investigated a suite of passive sensor-based vegetation and pothole detectors, but these approaches are often application specific, suffer from limitations specific to their modality, or require expensive, tailormade sensors. We present Augmentations for Vegetation, Obstacle, and Indentation Detection and Segmentation (AVOIDS), a perception system that combines approaches used by both vegetation and void detection algorithms to provide greater performance and robustness to data quality. Furthermore, we synthesize our detections with ranging information that can be used by downstream autonomous navigation behaviors. AVOIDS is a platform-agnostic solution that utilizes sensors that have utility beyond its specific application.