Viewing Motion Planning as Disassembly: A Decomposition-Based Approach for Non-Stationary Robots

An efficient single-query motion planning approach for non-stationary robots is presented. The approach exploits workspace information to identify highly constrained placements of the robot along a potential solution path. Due to the constraints imposed by the environment at these placements, we view them as an assembly, consisting of the robot and the environment. The original motion planning problem is then decomposed into a series of disassembly tasks. Each assembly is disassembled by moving the robot from the narrow passage into adjacent open regions. This disassembly can be performed efficiently by exploiting the geometric constraints imposed by the environment. To obtain a solution to the original motion planning problem, the resulting disassembly paths for two consecutive disassemblies along the solution path can be connected efficiently, as they meet in an open, unconstrained region of the workspace. Experimental evidence shows that the proposed approach results in a significant reduction in computational cost for motion planning in high-dimensional configuration spaces.