Cognitive Robotics Group

Research Overview

With the ongoing challenges of an aging population and labor shortages, there is increasing social demand for robotic technologies that can support or replace household tasks. To meet this demand, it is essential to develop robots with general and flexible motion generation capabilities that can handle complex and diverse tasks such as cooking and laundry.
The Cognitive Robotics Group is working on the development of robots equipped with human-like cognitive functions, using motion generation models inspired by the theory of predictive coding. This theory suggests that the human brain processes information by minimizing the error between sensory input and top-down predictions. Based on this idea, we construct deep predictive learning models that integrate bottom-up processing (sensory input) with top-down processing (intentions and predictions). These models enable robots to flexibly adapt their actions to changes in objects and environments.
Our research focuses on tasks involving dynamic and deformable objects, such as manipulating clothes with changing shapes or handling ingredients whose states change during cooking. In such tasks, the robot must adjust its behavior based on visual and physical properties, including shape, color, and temperature.
Current research topics include:
-adaptive control of the robot’s visual attention,
-decision-making under uncertainty in predicted future actions, and
-generation of coordinated bimanual movements.
This research is conducted in collaboration with the Motion Generation Group of Professor Tetsuya Ogata’s Laboratory in the Department of Intermedia Art and Science, School of Fundamental Science and Engineering.