Fast and Memory-Efficient Planning in C-space: Modified Bi-directional RRT* Algorithm for Humanoid Robots

In this paper we address the problem of faster and memory-efficient path planning for anthropomorphic manipulators with multi-link collision avoidance. As a solution, we present a modification of the intelligent bi-directional RRT* algorithm working in a C-space, where we don’t generate excess vertices of a tree by estimating their locations. Suggested algorithm is validated for grasping task with iCub humanoid robot. Comparison with others RRT* modifications demonstrate that we find the similar-length path with the significant improvement in a planning time, reduced amount of the memory to be allocated and show 100% success rate for the cases, where others planners will likely fail.