Home // UBICOMM 2024, The Eighteenth International Conference on Mobile Ubiquitous Computing, Systems, Services and Technologies // View article

Attraction-Based Reinforcement Learning: A Real-Time Approach Using Techniques Based on the Animal Behavior

Authors:
Marcos Laia
Márcio Mendonça
Edimilson Santos
Rodrigo Palácios
Wesley Guimarães
Janaina Gonçalves

Keywords: Unscented Kalman Filter; Animal Behavior; Machine Learning; Autonomous Systems.

Abstract:
Building a virtual world with simulated physical phenomena based on attraction and repulsion rules offers a unique opportunity to move agents according to these rules and plan actions mimicking real-world animal behavior. This paper presents an attraction-based algorithm using the Unscented Kalman Filter (UKF) to learn and predict opponent behavior in real time. The algorithm leverages attraction and repulsion forces to simulate physical interactions, facilitating robust predictions and learning accurately. Agents can optimize their strategies through reinforcement learning by adjusting attraction and repulsion parameters. Our results demonstrate the algorithm's effectiveness in dynamic environments, compared with traditional Q-learning methods, especially in low-frame conditions.

Pages: 10 to 15

Copyright: Copyright (c) IARIA, 2024

Publication date: September 29, 2024

Published in: conference

ISSN: 2308-4278

ISBN: 978-1-68558-191-6

Location: Venice, Italy

Dates: from September 29, 2024 to October 3, 2024