Graph Based Model Predictive Control of a Planar Bipedal Robot

摘要

In this paper, a new control method for a planar bipedal robot, which we call here Graph-based Model Predictive Control, is proposed. This method consists of two phases: the graph construction phase and the realtime control phase. In the graph construction phase, a directed graph on the state space of the control target is constructed off line. In the realtime control phase, the controller drives the state of the control target so as to make it move through graph nodes connected by directed edges. By tracing directed edges, a model predictive control is achieved in some sense. Moreover, since the directed graph is constructed in advance, the realtime computational cost is dramatically reduced compared with the ordinary MPC. In addition, by constructing multiple directed graphs based on different cost functions, one can design multiple motions and switching trajectories among them in a uniform way. The proposed method is applied to the speed changing control problem of a bipedal walker on a 2-dimensional plane and its effectiveness is verified by numerical simulation.