Recognition of manipulated objects by motor learning with modular architecture networks
✍ Scribed by Hiroaki Gomi; Mitsuo Kawato
- Publisher
- Elsevier Science
- Year
- 1993
- Tongue
- English
- Weight
- 959 KB
- Volume
- 6
- Category
- Article
- ISSN
- 0893-6080
No coin nor oath required. For personal study only.
✦ Synopsis
For recognition and control of multiple manipulated objects, we present two learning schemes for neuralnetwork controllers based on fi, edback-error-learning and modular architecture. In both schemes, the network consists of a recognition network and modular control networks. In the first scheme, a Gating Network is trained to acquire object-specfw representations Jbr recognition of a number o[objects (or sets of objects). In the second scheme, an Estimation Network is trained to acquire.[itnction-speco~c, rather than object-specftc, representations which directly estimate ph)'sical parameters. Both recognition net~;z~rks are trained to ident~[.i, manipulated objects usit~e somatic and~or visual in./brmation. After learning, appropriate motor commands./br manipulation of each object are issued by the control netu'orks which have a modular structure. By simulation of simple examples, the potential advantages and disadvantages o.f the two schemes are examined.