Online path planning in dynamic environments using the curve shortening flow method

摘要

Handling tasks with robots which are interacting in a shared workspace provide a high risk of collision. A new approach based on curve shortening flows enables collision-free path planning for robots within a varying environment on basis of a workspace model. Thereby a global path planning method based on geometrical curvature flow is combined with the locally and reactively acting potential field method. In the authors' previous work this path was formulated as a non-linear partial differential equation and solved using explicit time-integration. This article presents a linearized model of this method which overcomes the numerical stability problems of the former formulation and allows a more efficient computation which is strongly required for the algorithm to be run on a realtime robot control. After a detailed explanation of the algorithm simulative and experimental results using the example of a handling task performed by two 5-axis manipulators is presented and effects of different parameter settings on the resulting path are discussed.