Back to Search Start Over

Hybrid Bio Inspired-Based Optimized Neural Network for Real-Time Evasion of Multi-Robot Systems in Dynamic Environments

Authors :
Alahmari, Saad
Salameh, Anas A.
Innab, Nisreen
Deebani, Wejdan
Alhomayani, Fahad M.
Shutaywi, Meshal
Ghoneim, Mohamed E.
Source :
Dynamic Games and Applications; 20240101, Issue: Preprints p1-18, 18p
Publication Year :
2024

Abstract

The immediate shift in the environment leads to many challenges in pursuit-evasion studies. Existing techniques need help to handle these unexpected changing scenarios. To address these situations, we proposed a hybrid solution called BINNQN. Hybrid Bio-Inspired Neural Network combined with the advantage of Deep Q-Network (DQN) to improve the performance of multi-robot evasion in unpredictable environments. Here, the BINN features help to simulate neurodynamic activities that approach a pursuit-evasion scenario. It also allows the immediate, real-time development of collision-free evasive paths. On the other hand, the DQN component provides continuous knowledge improvement to address the local minimal concerns and help adapt it to complex, unexpected risks. Additionally, DQN learns from the reward system to evaluate and select the best evasive actions based on continuous feedback from the environment; this helps the system increase the evasion time and decrease the risks of capture and collision. This hybrid approach involves the BINN-based quick response ability combined with the adaptive learning and decision-making ability of DQN. The existing studies implement their proposed approach with real-world mobile robots. Still, in contrast, this study creates an app to test the proposed hybrid BINNQN scenario, which provides a safe and cost-effective platform for testing and visualizing complex multi-robot evasion techniques without the risks of real-world hardware involvement. To encourage the user’s accessibility and their interaction knowledge with the model, we planned to execute the proposed model using the mobile app. Combining the advanced techniques of BINN and DQN, the proposed BINNQN model works well and effectively contributes to this field. The existing BINN achieves the stable value of 21.2, whereas the proposed hybrid achieves a stable state of 23.4, which is a promising approach for real-time evasion in complex multi-robot systems environments.

Details

Language :
English
ISSN :
21530785 and 21530793
Issue :
Preprints
Database :
Supplemental Index
Journal :
Dynamic Games and Applications
Publication Type :
Periodical
Accession number :
ejs68300030
Full Text :
https://doi.org/10.1007/s13235-024-00608-0