Algorithm evaluation demonstrates up to 99.98% data reduction in multi-robot LiDAR map merging, highlighting feasible deployment on resource-constrained platforms.
Key Points
To develop a communication-efficient LiDAR map merging framework that significantly reduces inter-robot bandwidth consumption while maintaining mapping accuracy in resource-constrained environments.
Formulated selective data exchange as a three-stage cascaded optimization problem on an exchange graph where vertices represent keyframes and edges represent candidate inter-robot loops.
Optimized scans sequentially for overlap, balanced transmission cost, and geometric-perceptual quality to transmit only a minimal scan subset.
Evaluated performance across five public and four in-house datasets spanning cave, planetary-analog, indoor, and outdoor settings, including real-world non-line-of-sight communication tests.
Reduced data exchange volume by up to 5000×, achieving up to a 99.98% reduction (from 7000 MB down to 1.3 MB on the HeLiPR dataset) without loss of alignment accuracy.
Enabled real-time map merging on resource-constrained embedded platforms where conventional whole-scan and downsampling methods fail.
Maintained reliable map alignment under simulated network degradation and real-world non-line-of-sight conditions with packet dropout rates reaching 73.3%.