Takuya Sugimoto and Manabu Gouko
exploratory behavior, active perception, reinforcement learning, mobile robot
Discernment behavior is a type of exploratory behavior that supports object feature extraction. We proposed an active perception model that can learn discernment behavior using reinforcement learning. This model acquires an effective behavior that extracts features from objects. However, the learning of the exploratory behaviors takes a comparatively long time in this model. Additionally, this model often fails to acquire the exploratory behaviors. The cause of this kind of unstable learning is the influence of the past feature vectors. In this paper, we improved the learning method of the previous model. To verify the effectiveness of the improved model, we carried out a mobile robot simulation. The results indicate that our proposed model learns the exploratory behavior more stably than the previous model.
Important Links:
Go Back
