iRobotLab

研究/各研究の話/2012年度

強化学習におけるセンサの重要度に応じた効率的な意思決定(Efficiently Decision making in reinforcement learning based on importance of sensor)†

Summary†

• Focusing on a relationships between sensor value and reward as a barometer of importance of sensors.
• Calculating multiple regression coefficient between imput values of sensors and reward as degree of importance of sensors.
• To select an action, temporary Q-space is constructed from original Q-space based on degree of importance of sensors.
• Higher degree of importance of sensors, more the number of state of temporary Q-space.

Conceptual diagram†