Occlusion-Aware Motion Planning of Autonomous Robots in Cluttered and Unknown Environments

Navigation through unknown, cluttered environments is a fundamental and challenging task for autonomous vehicles as they must deal with a myriad of obstacle configurations typically unknown a priori. Challenges arise because obstacles of unknown shapes and dimensions can create occlusions limiting sensor field of view and leading to uncertainty in motion planning. There have been limited studies on the topic of occlusion-based motion planning, and they are primarily centered around safety assurance under uncertainty. However, taking advantage of properties of these occlusions can allow for fast, agile navigation. The work presented in this thesis builds around this concept and proposes a framework which leverages such occlusions to quickly navigate cluttered, unknown environments. The proposed framework presents a novel occlusion-aware motion planner which provides agile exploration by estimating gaps in point cloud data and shadows in the field of view to generate waypoints for navigation. We extend this planner to navigate quickly to a predefined goal in cluttered, unknown environments. Our scheme also proposes a breadcrumbing technique to save states of interest during exploration that can be exploited in future missions. For the latter aspect we focus primarily on the generation of the minimum number of breadcrumbs that will increase coverage and visibility of an explored environment. Extensive simulations and experiment results on an unmanned ground vehicle (UGV) are demonstrated to validate the framework, showing improvements over traditional state of the art frontier-based exploration methods.