不連続な変化を伴う制御対象へのモンテカルロモデル予測制御の適用

—台車の衝突を考慮した台車型倒立振子の振り上げ高速化と安定制御—

抄録

This paper describes the application of Monte Carlo model predictive control (MCMPC) to systems with collision phenomena. In general, a physical quantity related to velocity discontinuously changes at the moment of a collision. Gradient-based model predictive control that requires gradient information of the cost function can not handle the collision phenomena directly. On the other hand, MCMPC can calculate local optimal solutions and stationary solutions needed each time as far as the forward simulation is feasible. Therefore, we examined the problem of swing up and stabilizing the cart type inverted pendulum with the collision of the cart. If the controller can properly handle the discontinuity, faster swing-up may be realized by exploiting impact force caused by the collision of a cart within the evaluation function. In this paper, we show the result of simulations and real experiments, and report that we can realize speeding up of swing-up using collision.