2A1-E08 GMMを用いた複数予測モデル型パーティクルフィルタによる意図推定

概要

論文の詳細を見る
This paper presents a new model to predict an operator's intention in man-machine system based on the multi-mode particle filter wherein a Gaussian mixture model (GMM) is embedded as the observation model. In the proposed model, each prediction model in the particle filter corresponds to each primitive motion generated by the operator. The operator's intention can be predicted by calculating the likelihood of each model in real-time. In addition, the advantage of using GMM is to be able to represent a nonlinear and/or nonstationary relationship between the state of the particle filter and the observation signal measured from the sensor. This advantage is more emphasized when the model is applied to the prediction of human complex behavior and when the observation order is less than that of dynamical state. The usefulness of the proposed model is demonstrated by applying to a manipulating behavior for the carrier machine.
一般社団法人日本機械学会の論文