高速なガウス過程回帰を用いた予測モデルに基づくサンプリングベース動作計画 Sampling-based Motion Planning with a Prediction Model using Fast Gaussian Process Regression
Recently, motion/path planning methods become popular and achieve practical applications such as autonomous car. Most of studies focuses on the planning based on a given mathematical model, but the development of data-driven system identification method is also crucial for practical applications, since the precise model of a control target is not always available in advance. In this research, we propose a motion planning method where a fast Gaussian process regression is used as a model of the control target, since Gaussian process regression is a powerful non-parametric method which is widely used in various practical and complicate applications. Thanks to Bayesian property of Gaussian process regression, our method can deal with uncertainty of the prediction. We apply our method to the control problem of simple-pendulum with non-linearity and achieve the swinging up task.
- 電気学会論文誌. C
電気学会論文誌. C 135(5), 526-533, 2015