Planning a dynamic trajectory via path finding in discretized phase space

Abstract

The control of dynamical systems under constraints on the the controlling force can require long term trajectory preplanning to avoid entering dead ends in phase space. To handle such problems in a flexible way we suggest to formulate them as a path finding problem on a suitably discretized version of the phase space of the system. This path finding problem can then be solved with a standard relaxation method in a parallel computation by exploiting a physical analogy: the target point in phase space is considered as the source of a diffusing substance whose concentration gradient serves as a local guide to the destination point.