不確実性を伴う物体操作を実現する動作生成手法

人生活支援を実現するマルチタスク型ロボットの開発(4)

抄録

One of the challenging issues in robotics is dealing with uncertainty. In this research, we propose a method for robust motion generation in unlearned environments, including extrapolation, by introducing Bayesians to conventional deterministic RNNs. The proposed method learns a probabilistic model from training data and generates situation-specific behaviors from the learned probabilistic model. We confirmed that the proposed method can robustly execute tasks even in unlearned environments by using a robot simulator. Furthermore, we show that the Recurrent Dropout can be used to reduce the dependence of the initial weights of neurons on generalization performance.