| 103C6 |
|---|
| Path and Speed Profile of Hand Reaching Movement that is Acquired by Reinforcement Learning |
| AUTHORS |
| Katsunari Shibata (Oita University) |
| ABSTRACT |
| Human hand reaching movement has been well investigated to know the mechanism of the motion learning in human and living things. It has been reported that the path is almost a straight line and the speed profile is almost bell-shaped in short distance reaching tasks. The author showed by a simulation that the path is close to a straight line and the speed profile is close to bell-shape with some exceptions when two-link arm is trained by reinforcement learning using a neural network with the input of visual signals, joint angles, and angular velocities. In this paper, the simulation was done on the similar conditions as the human experiment. The visual field became wider and the reaching distance became longer than the previous simulation. After learning, the paths were curved and the speed profiles are not close to bell-shape and some speed profiles have two peaks against expectation. Especially, when the joint 2 was extended at the initial state, it began to be bent at first even if the hand could reach the target faster by keeping the joint to be extended. The reason has not been investigated yet. |
Back to Program
Time Table
Top of This Site
Annual Conference 2002