Adaptive Control of Visually Guided Grasping in Neural Networks
Abstract
We present a theory and prototype of a neural controller called INFANT that learns sensory-motor coordination from its own experience. INFANT adapts to unforseen changes in the geometry of the physical motor system and to the location, orientation, shape and size of objects. It can learn to accurately grasp an enlongated object without any information about the geometry of the physical sensory-motor system. This new neural controller relies on the self- consistency between sensory and motor signals to achieve unsupervised learning. It is designed to be generalized for coordinating any number of sensory inputs with limbs of any number of joints. The simulation model was implemented with real targets and movements using two stereo TV cameras and a multijoint manipulator. The theory of sensory-motor coordination was extended from single movements to movement sequences. The neural network controller in the proposed study has a number of application benefits. The controller will deal effectively in novel working environments such as in space because of its ability to deal with unforeseen changes in the mechanical plant and actuators. Its adaptability will allow continuous self-calibration and its genetic design will allow it to be implemented in many different robots. The parallel feedforward control architecture will make robot control very fast and the overlapping modifiable neural weights will allow fault tolerance. This will greatly reduce tooling costs, setup time and failure in unforeseen environments. Keywords: Computerized simulation; Artificial intelligence.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 12, 1990
- Accession Number
- ADA221387
Entities
People
- Michael Kuperstein