| 研究課題/領域番号 |
18K18130
|
|---|
| 研究機関 | 東京工業大学 |
|---|
研究代表者 |
高木 敦士 東京工業大学, 科学技術創成研究院, 特任助教 (70802362)
|
|---|
| 研究期間 (年度) |
2018-04-01 – 2021-03-31
|
| キーワード | grasp force / movement precision / force control / trajectory control / handheld tooling |
|---|
| 研究実績の概要 |
Using a new measurement methodology of measuring the power grasp force, we have succeeded in analyzing the adaptation of the human's control of precision, tooling force and the tool's trajectory during handheld tooling. We found that the rates of learning for each type of control was different, with tooling force decreasing most rapidly. Precision and the tool's maximum speed increased, but at a slower rate than the tooling force. These findings are important as it enables us to understand that tooling force is learned via fast supervised learning, whilst precision and trajectory control are learned using a slower unsupervised learning.
The findings are also novel as the ability to measure the control of movement precision have not been applicable to tasks like tooling. Previous methodologies required a robot to perturb the position of the human's hand to measure precision control, but this cannot be done during tooling where contact forces are large and the movement is rapid. The grasp force enabled us to measure the adaptation of precision control even in these extenuating circumstances.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
The experiments have progressed smoothly, with interesting results and fascinating insights into the way humans control their movement precision, tooling force and trajectory during handheld tooling.
In particular, the development of a new measurement methodology to simultaneously capture the changes in tooling force, movement precision and the insertion speed is instrumental to understanding how humans learn to use tools. The new insights will pave the way for a robotic tooling algorithm that learns the key aspects necessary for contact tooling.
|
| 今後の研究の推進方策 |
From the experiments, we have understood that supervised learning of the tooling force enabled subjects to rapidly learn to reduce it. We also plan an additional experiment to understand if the grasp force can be given as feedback to improve the speed of learning to control precision.
The next phase is to build a computational model of the tooling task. Here, we will employ optimal control or reinforcement learning to mimic human motor learning to obtain a robotic controller that successfully completes the tooling task.
|
| 次年度使用額が生じた理由 |
Research expenditure was somewhat lower this FY than expected, as the purchase of new hardware for experiments has been delayed. In the next FY, when the robot algorithm is developed, the transferred sum will be used to purchase new electromyography sensors to measure the subject's muscular activity during handheld tooling, and a new robotic interface will be necessary to implement the robotic algorithm.
|
