SANS annual report 1994

Previous Up Next


Perception-Action based Systems

Perception-Action based Systems

An ANN may be regarded as a sub-system for classification, diagnosis, control etc. Such a sub-system can be a component in a larger structure, an artificial neural system (ANS). The ultimate challenge for an ANS designer is to construct a complete system capable of controlling a sophisticated perception-action system in a reasonably complex environment. This would imply providing high performance artificial perception, motor control, goal direction, and real-time action planning and decision making in an integrated, working system.

One good testbed for such a design can be referred to as an "artificial creature". This implies building or simulating a system with sensory and perceptual abilities, motor apparatus, learning capabilities, motivations, goals etc. This "artificial creature" moves around in an environment, while trying to survive and improve its performance there. A related area of ANS design involves coupling neural networks to mechanical systems to produce a desired locomotor movement, such as swimming or walking.

Subtopics

  • Control of walking
  • Experiments with a neuro-mechanical model
  • Some experiments with neuro-fuzzy control
  • Motivation based reinforcement learning in autonomous robots

  • Next