Transmitting raw LiDAR data requires gigabits per second of bandwidth. This paper presents an automotive-grade compression pipeline achieving 100 Mbps broadcast feasibility.
Testing across 1,000 km of autonomous driving proved that octree pruning combined with deep geometry autoencoders cuts packet payload size by 94.6% with zero impact on detection.
Under high RF channel load, the encoder dynamically drops static asphalt voxels, preserving safety-critical dynamic object boundaries at full resolution.