Motor learning of body-tool-environment systems

We use many kinds of tools to achieve various tasks in daily life. For example, when we play baseball, we use a bat to hit a ball. Then the arm posture and hand position holding the bat are changed depending on the task, e.g. hitting a ball as far as possible or taking the bunting position. As seen from this, the manipulation of tools is strongly task-oriented. The present paper proposes a new learning method based on reinforcement learning, which can simultaneously obtain several hand positions holding the tool and create motion patterns to reach a goal. Several simulation results are shown.