Unifying model-based programming and randomized path planning through optimal search

Abstract

The deployment of robots at the World Trade Center (WTC) site after September 11, 2001, highlighted the potential for robots to aid in search and rescue missions that pose great threats and challenges to humans. However, robots that are tele-operated and tethered for power and communication are restricted in terms of their operational area. Thus, rescue robots must be equipped with onboard autonomy that enables them to select feasible plans on their own, within their physical and computational limitations. There are three main characteristics that a rescue robot's onboard system must posses. First, the system must be able to generate plans for mobile systems, that is, plans with activities and paths. Second, in order to operate as efficiently as possible, particularly in emergency situations, the system must be globally optimal. Third, the system must be able to generate plans quickly. This thesis introduces a novel autonomous control system that interleaves methods for spatial and activity planning, by merging model-based programming with roadmap-based path planning. The primary contributions are threefold. The first contribution is a model that represents possible mission strategies with activities that have cost and are constrained to a location. The second is an optimal pre-planner that reasons through the possible mission strategies in order to quickly find the optimal feasible strategy. The third contribution is a unified, global activity and path planning system. The system unifies the optimal pre-planner with a randomized roadmap-based path planner, in order to find the optimal feasible strategy to achieve a mission. The impact of these contributions is highlighted in the context of an urban search and rescue (USAR) mission.