AUTONOMOUS OPERATIONS AND SYSTEM INTEGRATION OF LOW-COST ROBOTIC VEHICLES WITH APPLICATION TO SMALL GROUND VEHICLES AND NANOSATELLITES
Johns Hopkins University
The goal of this thesis is to enable basic autonomous navigation capability of low-cost robotic vehicles, in situations where only an approximate model of the vehicle and environment is available and only high-level guidance from a human operator is expected. We develop a common framework that utilizes inertial and depth sensing to enable basic perception and employ standard planning and model-based control to enable autonomous navigation to a given goal state. The framework is applied to two domains: small satellites performing on-orbit navigation (using a mock-up environment and engineering models), and small ground vehicles traversing unpaved terrains (e.g. in a campus-like environment). For the satellite domain, our goal is to enable autonomous relative navigation between multiple spacecraft, that is needed for future missions that require e.g. segmented mirror on-orbit assembly, motivated by the fact that performing such missions using human-guided ground control is infeasible. For the small ground vehicle domain, unlike standard autonomous cars, our goal is to enable traversal of environments that do not necessarily have paved roads. We therefore develop algorithms that employ learning-based dynamical models which can adapt to more than one terrain. This thesis investigates both of these systems equipped with minimal sensing and actuation, and performs a feasibility analysis of the effectiveness of autonomous navigation.
nanosatellites, CubeSats, Optimal Control, Planning Algorithms, UGV, unmanned ground vehicles, global planning, local planning