UC Berkeley

In the past decade, the level of autonomy of unmanned vehicles has been rising rapidly from remote-controlled towards fully autonomous. Without human operators on board, teams of autonomous vehicles are the best candidates for high risk applications such as search and rescue after disasters and information gathering in hostile environments. For a team of autonomous vehicles to operate effectively in these scenarios, it must be able to respond promptly to environmental hazards and/or hostile entities. In this dissertation, a collaborative team evasion framework is proposed to maximize the survival time of a team of autonomous vehicles against a faster and more agile hostile agent. The proposed framework is based on an open-loop formulation of the single-pursuer-multiple-evader pursuit-evasion game that is conservative to the evaders and provides guarantees on team survival time in the worst-case scenario. An iterative open-loop approach that repeatedly solves the open-loop problem corresponding to the most current state of the game is developed to relax the conservatism of the open-loop formulation and enhance the survival time performance. Extensions to the framework make it possible to take into account the turning rate constraints of the evaders and uncertainties in the position of the pursuer. Numerical approximations are also proposed to reduced the required computation time. Through extensive simulations, the proposed framework is shown to produce reliable strategies for the evaders that result in significantly longer team survival time than previous work in the literature.