A Collaborative Real-Time Object Detection and Data Association Framework for Autonomous Robots Using Federated Graph Neural Network

Autonomous robotics require secure and decentralized decision-making systems that ensure data privacy and computational efficiency, especially in critical areas. Current
centralized models or human input are associated with data breaches and security vulnerabilities. To counter these, we propose CoRODDA, a dedicated framework combining federated learning and graph neural networks. It enhances object detection and data association in autonomous robots, enabling them to learn from local data while preserving privacy and interpreting graph-structured associated data to understand their environment. The experiments show the effectiveness of CoRODDA compared to the state-of-the-art, particularly in non-detected objects, improving data privacy and decision-making capabilities