MR-SLAM brings multi-robot SLAM supervision into mixed reality with Quest 3

MR-SLAM addresses a core problem in multi-robot mapping: operators typically monitor fleets through a flat RViz window, losing spatial context as the number of robots grows. To solve this, researchers from the MM-SpatialAI workshop at ICRA 2026 built a mixed-reality supervision system using a Meta Quest 3 in passthrough mode. The system simulates three TurtleBot3 robots in Unity 2022 LTS, each publishing a simulated 2D LiDAR scan (180 rays over a 90-degree arc at 10 Hz) to ROS 2 running on a standard Ubuntu laptop (i5, 16 GB). Spatially anchored dashboard panels show per-robot coverage, scan rate, map dimensions, and latency directly in the operator’s environment. Spatial occlusion via Meta's MRUK hides robots behind real furniture, adding realism.

The system’s performance was validated across five 9-minute sessions, delivering 8.83 Hz scan delivery, 94.7% cross-instance occupancy consistency between robot pairs, and 6.3 ms median transform jitter, mapping up to 26.7 m² of a 41 m² grid. Integration challenges included TF conflicts (solved by having Unity publish only the odom->base_footprint subtree while SLAM Toolbox owns map->odom) and a QoS mismatch between ros_tcp_connector’s RELIABLE clock and SLAM Toolbox’s BEST_EFFORT subscription. The paper is presented as a reference implementation on consumer hardware, not a finished product. Open questions remain about maintaining alignment between spatial anchors and the ROS map frame over long sessions, and moving from simulated to physical robots. The full code is available on GitHub.