A NOVEL FRAMEWORK TO EFFICIENT PATH PLANNING THROUGH REAL-TIME COST MAP GENERATION USING NEURAL NETWORKS FOR SEARCH AND RESCUE MISSIONS.

Abstract

Search and rescue (SAR) missions are critical operations that demand swift and efficient execution to save lives in the aftermath of dis- asters. This paper introduces a novel framework for optimizing path planning in robotic SAR missions through the generation of real-time cost maps using neural networks. Our approach integrates static topo- logical data with dynamic mission findings to create an amalgamated cost map that prioritizes urgent and accessible regions. We propose a modified U-Net architecture, specifically adapted for SAR applications, which enables adaptive cost prediction and enhances learning capabilities in complex, evolving environments. Extensive simulations demonstrate significant improvements in survivor location efficiency compared to traditional baseline approaches. The framework’s ability to continuously update based on real-time data ensures robust adapt- ability to the dynamic nature of SAR missions. By bridging the gap between theoretical models and practical implementation, our method has the potential to revolutionize crisis response strategies, offering a more agile and effective approach to robotic search and rescue operations.

DOI: 10.7176/CEIS/15-1-10

Publication date: October 30^th 2024