Toward autonomous cleaning with manipulator arms
Cao, Junmei · ANU Open Research (Australian National University) · 1999
This thesis is to develop techniques to enable a manipulator arm to perform a cleaning task in an unknown obstacle-filled environment using only force sensing.Our investigation focusses on reactive control, graph-based modelling for manipu lator arms and graph-based motion-planning algorithms.Reactive control strategies have produced significant results in improving the ability of wheeled and legged mobile robots to cope with unfamiliar environments and uncer tainties.However, it seems that very few attempts have been made to apply these strategies to manipulator arms.In this thesis, we extended the purely reactive con troller [10] with the ability to plan, and used only force sensing.The controller was applied to an Industrial Scientific SCARA robot to explore an unknown obstacle-filled environment.The task is to cover all free space to simulate cleaning.The motion planning problem for manipulator arms is different from mobile robots.The arm kinematics allows each task position to be reached in a number of configura tions.Our graph-based modelling for manipulator shows manipulator motion planning problem can be formulated as a graph search.To work in unknown environments, robots have to re-plan motions on-the-fly as new information is acquired.In the thesis, we present a new algorithm to minimise re-planning computation on uniform graphs by incrementally repairing data.It is es sentially a simplified version of Stentz's D* algorithm [8], but, it is easier to implement, and more efficient for uniform graphs searches.We also investigate two graph-based motion planning algorithms in the context of sensor-based motion planning: an optimistic shortest path algorithm and a depth first search algorithm.We compare the DFS algorithm with the OSP and show that it achieves similar average-case performance if an original technique that we call "treeiii