Reactive task planning for multi-robot systems in partial known environment

The thesis investigates the planning and control problem for a group of mobile agents moving in a partially known workspace. A task will be assigned to each robot in the form of a linear temporal logic (LTL) formula. First an automaton-based method is introduced for the motion planning of a single agent, which guarantees the satisfaction of the assigned LTL task. Then a model-predictive controller considers state and input constraints leading the agent to a safe navigation. Based on a real scenario of a partial-known environment and agents can have only local sensing, two decentralized control strategies are proposed for online re-planning, which rely on a sampling-based algorithm. The first approach assumes local communication between agents, while the second one exploits a more general communication-free case. Finally, the human-in-the-loop scenario is considered, where a human may additionally take control of the agents, a mixed initiative controller is then implemented to prevent dangerous human behaviors while guarantee the satisfaction of the LTL specification. Using the developed ROS software package, several experiments were carried out to demonstrate the effectiveness and the potential applicability of the proposed strategies.