A reactive/deliberative planner using genetic algorithms on tactical primitives

摘要: Unmanned aerial systems are increasingly assisting and replacing humans on so-called dull, dirty, dangerous missions. In the future such will require higher levels of autonomy to effectively use their agile maneuvering capabilities highperformance weapons sensors in rapidly evolving, limited-communication combat situations. Most existing vehicle planning methods perform poorly realistic scenarios because they do not consider both continuous nonlinear system dynamics discrete actions choices. This thesis proposes a flexible framework for forming dynamically realistic, hybrid plans composed parametrized tactical primitives using genetic algorithms, which implicitly accommodate through fitness function. The combines deliberative with specially chosen react fast changes environment, as pop-up threats. Tactical encapsulate elements together, switchings define primitive type parameters capture stylistic variations. demonstrates combined reactive/deliberative problem involving two-dimensional navigation field threats while firing deploying countermeasures. It also explores planner’s performance respect computational resources, dimensionality, design, planner initialization. These explorations can guide further algorithm design autonomous tactics research. Thesis Supervisor: Christopher Dever Title: Senior Member, Technical Staff, C.S. Draper Laboratory John Deyst Professor Aeronautics Astronautics