Abstract

Recurrent networks are proposed for designing neurocontrollers in robotic control systems. An evolutionary algorithm is used to develop a neurocontroller for a robot clearing an arena by pushing boxes to the enclosing walls. The number of boxes that the robot pushes to the walls, within an allocated time, is the measure of its fitness. The initially feedforward networks evolve and produce high fitness neurocontrollers with 30% (and sometimes even 70%) fewer recurrent links compared to networks fully recurrent from the start.

Original languageEnglish (US)
Title of host publicationProceedings of the National Conference on Artificial Intelligence
Editors Anon
PublisherAAAI
Pages1378
Number of pages1
Volume2
StatePublished - 1996
EventProceedings of the 1996 13th National Conference on Artificial Intelligence. Part 2 (of 2) - Portland, OR, USA
Duration: Aug 4 1996Aug 8 1996

Other

OtherProceedings of the 1996 13th National Conference on Artificial Intelligence. Part 2 (of 2)
CityPortland, OR, USA
Period8/4/968/8/96

All Science Journal Classification (ASJC) codes

  • Software

Cite this