Motion Planning Algorithms For Mobile Manipulator In The Presence Of Obstacles

The main purpose of this thesis is to address the problem of indoor motion planningfor mobile manipulator in the presence of obstacles. The indoor environment we usedhere is obtained from a real indoor occupancy grid map, and is represented as a set ofpolygons. The model of Pioneer3DX is used as the mobile manipulator system.Though rapidly-exploring random tree (RRT) is the most promising motion planningalgorithm in practice use, and it is pretty suited for dealing with differential constraints,the simulation results indicate that directly implementing RRT in our constructedenvironment is inefficient, as indoor environment is usually divided into rooms by walls,doorways and passageways, and tasks may refer to crossing several rooms, any greedy orbiased strategies will inevitable result in exploring towards bad directions.Two novel methods have been presented to overcome this problem. The first one(Topological-RRT) tries to construct a topological map to guide the local sampling, themajor contribution of this method is the process of obtaining this topological map, that is,employ image processing technique and flood fill algorithm to segment the topologicalnodes and detect the connections between them. The second one (Zone-RRT) takesadvantage of big empty space (zone) to guide the bottom planning, where a zone means acollision free rectangle empty space, and zone graph is constructed first to find a feasiblepath in the global layer, and then modified biased-RRT is employed to perform withinthese adjacent zones. A critical difference of this method over related workspaceinformation guided methods is that they rely on the environment decomposition whileZone-RRT employs random sampling and collision detection techniques to fulfill theguiding work. Simulation results suggest that both of these two methods offer significantimprovements in performances.