Connectionist approaches to robot path finding
The first part of the report presents the robot path finding problem. It illustrates how human beings solve this problem in two stages. Knowledge-based approaches should be confined to the first stage of processing, while connectionist approaches should be used to address the second stage. The second part outlines the drawbacks of classical approaches to robot path finding and points out possible ways of circumventing some of them through connectionist approaches. It also identifies classifier connectionist path-finders as the most suitable systems for handling the second stage, and constructive connectionist path-finders as systems somewhere between planning and classifier path-finders. In the third part, the existing connectionist path-finders are classified and their advantages and shortcomings are highlighted. At the top level the distinction is made between constructive and classifier path-finders, then each category is broken down according to the underlying learning principle of its members, if any. Finally, the fourth part suggests a way to combine knowledge-based and connectionist techniques.
Bibliographic Reference: Extract: Progress in Neural Networks Series, Vol. 3 (1990)
Availability: Ablex Publishing Corporation, Norwood, New Jersey (US)
Record Number: 199110758 / Last updated on: 1994-12-02
Original language: en
Available languages: en