Adaptive Sampling-based Motion Planning of Articulated Robots

Abstract

This thesis handles with a motion planning of high DOF (degree-of-freedom) articulated robots. For this objective, we propose the sampling-based path planning method, an adaptive RRT (Rapidly-exploring Random Tree) algorithm. Here the adaptive body selection method selects the bodies or the active joints of the robot depending on the complexity of the path planning. That is, it determines the bodies nearby the workspace obstacle and then adds more bodies if it failed to the obstacle avoidance before. So the joints corresponding to the selected bodies are used for the path planning. This means that the configuration space with reduced-dimension can be constructed. For each random sampling, the adaptive RRT algorithm tries to extend the tree in the configuration space with reduced-dimension. Therefore comparing with the path planning in the full-dimensional configuration space, the proposed algorithm guarantees the reduced planning time and makes possible to the robot to use less number of joints in the path planning. The path generated by the adaptive RRT algorithm is needed to be transformed to a smooth trajectory by using the trajectory pruning algorithm developed in this thesis. Generally the initial path from the sampling-based path planner is not only unnecessary long but also jerky. This path is not time-efficient and may induce a sudden change of the velocity and acceleration to the robot. The aim of the trajectory pruning is to solve this problem. This algorithm removes the redundant path nodes from the initial path and then connects the remaining nodes using the B-spline curve. Even though the initial path is collision-free, the path is smoothed by the different shape trajectory. Thus the algorithm repeatedly tries to eliminate the path node so that the updated B-spline curve is collision-free. Then the result curve is used for the robot trajectory so the reduction of the path length and the constraints on velocity/acceleration are satisfied simultaneously. To verify the performance of the proposed algorithms, we have implemented these for the motion planning of dual-arm robot. We tested the algorithm though various kinds of path planning scenarios. It is shown that the propose algorithms can find a path within shorter time comparing with the original RRT algorithm. And for a real application of robot motion planning, we also describe the results on the motion planning of two dual-arm robots for the cell-phone packaging. By planning trajectories that the dual-arm robot picks the cell-phone part and place it to the cell-phone box, we shows the our algorithms can be applied to robotic automation.