MOTION AND TASK PLANNING FOR UNMANNED MULTICOPTER SYSTEMS

Abstract

This thesis aims to investigate techniques in motion and task planning that increase the autonomous capability of the unmanned multicopter aerial systems. For motion planning, we propose to parameterize the motions by their boundary conditions and introduce the boundary state constrained primitives (BSCPs). We first integrate BSCPs into a model predictive control (MPC) framework for fast and efficient local motion planning. We then extend the usage of BSCPs into a sampling-based planning framework to generate a long and optimal trajectory. Rather than reaching a single target, complex task elements often involve temporal logic constraints. We investigate both mixed-integer programming method and automaton-guided method to compute safe trajectories under temporal task specifications. Finally, we propose a behavior tree based task interface that allows users to combine various existing task elements into a more complex mission, which is highly modular and intuitive for human operators. The motion and task planning systems proposed in this thesis have been implemented on real aerial vehicles and tested with both simulations and flight experiments.