1992 Fiscal Year Final Research Report Summary
Visual Recognition of Objects and Control of Hand Shaping in Grasping Movements.
| Project/Area Number |
03650338
|
|---|
| Research Category |
Grant-in-Aid for General Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
計測・制御工学
|
|---|
| Research Institution | The University of Tokyo |
|---|
Principal Investigator |
SUZUKI Ryoji Univ.of Tokyo, Dept.of M.E.I.P., Prof., 工学部, 教授 (80013811)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
UNO Yoji ATR H.I.P.R.L., Senior Researcher, ティ・アール人間情報通信研究所, 主任研究員 (10203572)
NISHII Jun Univ.of Tokyo, Dept.of M.E.I.P., Res. Ass., 工学部, 助手 (00242040)
|
|---|
| Project Period (FY) |
1991 – 1992
|
| Keywords | Grasping Movement / Neural Network / Sensory Integration / Data Compression / Data Glove |
|---|
| Research Abstract |
The brain must solve two important problems in grasping movements. The first problem concerns the recognition of grasped objects. Specifically, how does the brain integrated visual and motor information on a grapsed object? The second problem concerns hand shape planning. In other words, how does the brain determine the suitable hand posture according to the shape of an object and the task?
A neural network model which solves such problems has been developed. The network consists of multilayers of neurons with forward connections. The operations of the neural network are divided into the learning phase and the pattern generating phase. In the learning phase, internal representations of grasped objects are formed in the middle layer of the network by integrating visual and somatosensory information. In the pattern generating phase, the finger configuration for grasping an object is determined by using the relaxation computation of the network.
A neural network model for acquiring an internal representation of the weight of grasped objects was also developed. This model consists of two networks; the one calculates weight of the object from joint angles and torques and the other calculates torques from the weight and joint angles. The multilayered network combining these two networks can be used to learn arm movements without knowing the weight of the grasped object.
|
