Edge-Assisted Multi-Robot Visual-Inertial SLAM with Efficient Communication

A team of researchers led by Xin Liu and Shuhuan Wen has published a paper detailing a breakthrough in multi-robot navigation. Their system, an Edge-Assisted Multi-Robot Visual-Inertial SLAM framework, tackles the critical bottleneck of bandwidth in robot swarms. By implementing a novel robot-edge-cloud layered architecture, the system intelligently distributes the computational heavy lifting of Simultaneous Localization and Mapping (SLAM). The edge layer handles local processing, while the cloud ensures global map consistency, overcoming the limitations of a robot's onboard computer.

The core innovation is a highly efficient communication protocol. Instead of streaming raw sensor data, each robot transmits only losslessly compressed visual feature points and keyframe descriptors. This is paired with a 'lightweight SLAM method of optical flow tracking based on pyramid IMU prediction' to further reduce local computation. The result is a drastic reduction in the data required for real-time, collaborative mapping without sacrificing the precision needed for accurate navigation.

Validation on the standard EuRoC MAV dataset shows the system's practical superiority. It achieves the same or better positioning accuracy than current state-of-the-art centralized multi-robot SLAM schemes, but does so under a significantly lower computational load. Furthermore, it outperforms existing feature compression methods by transmitting a lower data volume. This efficiency is the key to enabling scalable, real-time operations for robot teams in environments with constrained communication, like search-and-rescue or warehouse logistics.