Biologically Inspired Architectures for Learning and Control – Part I - An Autonomous Systems Perspective

Authors
  1. Cheng, D.X.P.
Corporate Authors
Defence R&D Canada - Suffield, Ralston ALTA (CAN)
Abstract
This report reviews the state of the art in biologically inspired learning architectures for autonomous systems, which enable them to improve their performance on given tasks with practice. Learning requires some evaluation of the performance and feedback to the learner, enabling it to modify its strategy for performing the task in question. In view of the feedback involved, the study of learning is highly relevant to a study of robot control. In fact, this aspect of learning is frequently referred to as a control policy. Following an overview of the major issues in robot learning and the distinction between it and machine learning in artificial intelligence, this report discusses reinforcement learning, neural-network-based learning, and learning using evolutionary algorithms. The concluding section discusses some of the limitations of cognitive learning processes in real world situations and directions for future development.

Il y a un résumé en français ici.

Report Number
DRDC-SUFFIELD-TM-2006-235 — Technical Memorandum
Date of publication
01 Dec 2006
Number of Pages
36
DSTKIM No
CA031398
CANDIS No
530295
Format(s):
Electronic Document(PDF)

Permanent link

Document 1 of 1

Date modified: