Collaborative SLAM (ROMAN)¶

To demonstrate HERCULES as a benchmark sandbox, we provide a ready-to-run collaborative SLAM evaluation using ROMAN (MIT-ACL) on a City Block sequence with two UAVs and two UGVs.

Each robot runs LIO-SAM odometry with an open-set ROMAN object map; inter-robot loop closures are registered pairwise via CLIPPER.

Live mapping¶

Per-robot live trajectory, camera pose, and object map:

UAV 1 UAV 2 UGV 1 UGV 2

Final maps¶

Per-robot LIO-SAM final map with the ROMAN object map:

UAV 1 final map UGV 1 final map

Loop-closure alignment¶

ROMAN loop closure aligning UGV 1 and UGV 2 submaps pairwise via CLIPPER:

Loop-closure alignment

Running it¶

The evaluation runs on data exported by the HERCULES dataset pipeline and consumed through the optimized ROS 2 wrapper. See the dataset page for the sequence formats (ROS 2 bags / KITTI-style layouts).