CoSAR: Multi-Robot Collaborative Semantic Mapping over Wireless Networks

Abstract

The recent advancements in 3D-rich media interactive technologies, including Augmented Reality (AR) and Extended Reality (XR), have brought about a significant proliferation of applications in various industries. In response to this trend, multi-robot collaborative semantic mapping has emerged as a promising approach to improve the accuracy and speed of large-scale environment perception for intelligent robotic systems. This paper proposes a framework that utilizes cooperative self-adaptation-based wireless communication between the server and robotic agents (CoSAR) to enhance the efficiency of semantic mapping in large-scale environments. Integrating this adaptive wireless communication reduces the time required for mapping, and each robot’s CPU and memory utilization is effectively decreased. Simultaneously, dynamic object removal and global map optimization are executed based on semantic segmentation information, resulting in enhanced accuracy and robustness of the semantic octo-tree map. The framework has been validated and tested on publicly available datasets using network simulations and deployed on real robots for real-life operations.