Towards A UAV-UGV Robust Exploration System

Abstract

This thesis investigates a heterogeneous robotic team comprising an Unmanned Ground Vehicle (UGV) and an Unmanned Aerial Vehicle (UAV) operating in unknown or partially known environments under time and safety constraints. Heterogeneous UGV-UAV systems are ideal for navigating \textit{a priori} unknown or partially unknown environments: utilizing the UAV's birds-eye perspective to map the environment or guide the UGV through obstacle-ridden terrain. The primary focus of this work is guaranteeing the UAV's safe deployment to a set of waypoints and return to the UGV while considering power constraints and communication limitations. Our proposed solution consists of an offline closed-loop pursuit evasion differential game and an online trajectory planner. The offline rendezvous game conducts a worst case analysis to determines an invariant safe set using the Hamilton-Jacobi-Isaacs variational inequality. The online framework then approximates the UGV's forward reachable set (FRS), and deploys the UAV if all states intersect with the capture basin. We solve variations of the suicidal pedestrian differential game, using numerical and deep-reinforcement learning Hamilton Jacobi solvers, to determine trade-offs between generality and over-approximation when simulating the online planner.