確率ロボティクス |
Probabilistic Robotics |
確率的要因の存在するサーボシステムの状態量推定と,その推定を利用して動作する制御系の構成方法について研究しています.
特に最近,非ガウス・非線形モデルの確率密度関数の推定が可能なパーティクルフィ
ルタを状態観測機として制御系に組み込む方法論を提案しました.
Since the particle filter is able to apply to
non-Gausian and nonlinear system models, it is capable of wide
application than the Kalman filters.
In this paper, a construction method of a state feedback
control system using the particle filter as an observer for
a probabilistic state estimation is described.
In order to be robust to non-Gaussian noise, maximum a posteriori
probability estimation extraction method and evaluation method of the
effective sample size have been incorporated in the
particle filter.
Then, effectiveness of the constructed system is verified
experimentally, and the effectiveness of the state observer constructed
by the particle filter is indicated by comparison with the Kalman filter.
A state feedback control system including the particle filter state observer. | Time evolution of the estimation. The probability distribution of the measurement noise follows two Gaussian distributions. |