SHARE: Towards Head-Mounted AR with User-Centric SLAM in Shared Human–Robot Workspaces

In shared human-robot collaboration (HRC) environments, existing multi-agent Simultaneous Localization and Mapping (SLAM) systems often treat all agents (robots and AR headsets) as homogeneous. This uniform resource allocation causes high latency for AR users, who require sub-20ms motion-to-photon response times to prevent visual discomfort, while robots are significantly more tolerant of latency. The authors ask: Can we reduce latency for AR users in multi-agent SLAM systems by exploiting the heterogeneous latency requirements of different agents?

The authors developed SHARE, a system that implements a user-centric approach to multi-agent SLAM. First, they built a Quality-of-Experience (QoE) model that assigns different priority levels to agents based on their specific latency tolerance. Second, they implemented a Proportional-Integral-Derivative (PID) scheduler that dynamically adjusts transmission priorities based on real-time environmental conditions, such as the number of visual features available. Finally, they introduced a visual overlap calculation method that identifies and trims redundant map merging operations by predicting merge utility based solely on agent pose data, rather than transmitting full feature points.

SHARE was evaluated using Meta Quest 3 headsets and a TurtleBot 4 robot. The system achieved an average latency of 13.22 ms for AR users, a 43.3% reduction compared to the baseline COVINS-G system, while maintaining sub-2-centimeter tracking accuracy. User studies confirmed that participants perceived significant improvements in robot responsiveness and movement predictability, with 65% of users preferring SHARE over existing solutions.

As AR becomes a standard interface for controlling robots in collaborative spaces, the "one-size-fits-all" approach to SLAM computation is no longer sufficient. By demonstrating that system resources can be intelligently reallocated based on the specific needs of the human user versus the robot, this work provides a scalable path toward seamless, low-latency AR-HRC interfaces.