D-Compress: Detail-Preserving LiDAR Range Image Compression for Real-Time Streaming on Resource-Constrained Robots
New method preserves critical geometric details for mapping and object detection while slashing data size.
A team of researchers has introduced D-Compress, a novel framework designed to solve a critical bottleneck in robotics: efficiently streaming high-fidelity LiDAR data from resource-constrained robots to edge servers. The core innovation addresses the shortcomings of using standard image/video codecs (like JPEG or H.264) on LiDAR range images. While convenient, these human-vision-optimized codecs often discard the precise geometric details essential for robotic perception, degrading the performance of crucial tasks like 3D mapping and object detection.
D-Compress tackles this by integrating specialized intra- and inter-frame prediction with an adaptive discrete wavelet transform approach for compressing prediction residuals, a method tailored to preserve geometric structure. Furthermore, the team developed a new rate-distortion optimization (RDO) algorithm specifically modeled for range image compression, enabling intelligent bitrate control under dynamic network bandwidth conditions. Extensive evaluations show D-Compress significantly outperforms state-of-the-art methods, maintaining superior geometric accuracy and downstream task performance even at compression ratios exceeding 100x, all while running in real-time on limited hardware.
The paper, set to appear at IEEE ICRA 2026, validates the framework's robustness in simulated dynamic bandwidth scenarios. This breakthrough is pivotal for enabling more complex, cloud-assisted robotic applications—from autonomous delivery to remote inspection—where reliable, high-quality perception data must be streamed from cheap, low-power robots operating in the field.
- Achieves over 100x compression ratios while preserving geometric details critical for robot perception tasks.
- Introduces a new RDO-based rate control algorithm specifically for LiDAR range images, enabling robust performance under dynamic bandwidth.
- Outperforms state-of-the-art compression methods in both accuracy and downstream task performance (e.g., object detection) and runs in real-time on constrained hardware.
Why It Matters
Enables cheap, low-power robots to stream high-quality 3D perception data in real-time, unlocking advanced cloud-assisted autonomy for logistics and inspection.