The purpose of this paper is to explore the effect of adding known amounts of memory to pure reactive systems in a variety of tasks. Usinga finite state machine approach, we construct controllers for a simulated robot for five tasks—obstacle avoidance, wall following, exploration, and box pushing—with two sensor configurations using evolutionary computation techniques, and compare the performance of stateless and memory-based controllers. For obstacle avoidance and exploration no significant difference is observed; for wall-following and box pushing, stateless controllers are significantly worse than memory-based but increasing amounts of memory give no significant increase in performance. The need for memory in these cases reflects a need to discriminate sensorimotor contexts to effectively perform the task.
|Title of host publication||Advances in Artificial Life - 6th European Conference, ECAL 2001, Proceedings|
|Editors||Jozef Kelemen, Petr Sosik|
|Number of pages||10|
|ISBN (Print)||3540425675, 9783540425670|
|Publication status||Published - 2001|
|Event||6th European Conference on Advances in Artificial Life, ECAL 2001 - Prague, Czech Republic|
Duration: 2001 Sept 10 → 2001 Sept 14
|Name||Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)|
|Other||6th European Conference on Advances in Artificial Life, ECAL 2001|
|Period||01/9/10 → 01/9/14|
Bibliographical notePublisher Copyright:
© Springer-Verlag Berlin Heidelberg 2001.
All Science Journal Classification (ASJC) codes
- Theoretical Computer Science
- Computer Science(all)