Research Analyzer
← Back ICRA 2026

Have We Scene It All? Scene Graph-Aware Deep Point Cloud Compression

Nikolaos Stathoulopoulos, Christoforos Kanellakis, George Nikolakopoulos

PDF

AI summary

Key figure (auto-extracted from paper)
A scene graph-conditioned deep autoencoder compresses 3D point clouds by up to 98% while preserving geometric and semantic fidelity for efficient multi-robot data transmission.
Point Cloud Compression Semantic Scene Graphs Deep Learning Multi-Robot Systems Edge Computing 3D Perception

Problem

The massive size and irregular structure of LiDAR point clouds strain bandwidth and storage, degrading performance in multi-agent robotic systems. Existing compression pipelines fail to leverage semantic or relational scene structure to guide efficient encoding.

Approach

The framework partitions point clouds into semantically coherent patches via a scene graph, encodes them with a FiLM-conditioned transformer into compact latent vectors, and reconstructs them using a bounding-box-guided folding decoder.

Key results

  • Up to 98% data reduction on SemanticKITTI and nuScenes
  • Outperforms Draco and MPEG codecs in compression metrics
  • Preserves geometric and semantic fidelity in reconstructed patches
  • Maintains raw-LiDAR-level accuracy for multi-robot pose graph optimization and map merging

Why it matters

Enables reliable, bandwidth-efficient 3D data sharing for edge-computing multi-robot systems without compromising navigation or perception tasks.

Abstract

Efficient transmission of 3D point cloud data is critical for advanced perception in centralized and decentral- ized multi-agent robotic systems, especially nowadays with the growing reliance on edge and cloud-based processing. However, the large and complex nature of point clouds creates challenges under bandwidth constraints and intermittent connectivity, often degrading system performance. We propose a deep compression framework based on semantic scene graphs. The method decom- poses point clouds into semantically coherent patches and encodes them into compact latent representations with semantic-aware encoders conditioned by Feature-wise Linear Modulation (FiLM). A folding-based decoder, guided by latent features and graph node attributes, enables structurally accurate reconstruction. Experiments on the SemanticKITTI and nuScenes datasets show that the framework achieves state-of-the-art compression rates, reducing data size by up to 98% while preserving both struc- tural and semantic fidelity. In addition, it supports downstream applications such as multi-robot pose graph optimization and map merging, achieving trajectory accuracy and map alignment comparable to those obtained with raw LiDAR scans.

Index terms

Range Sensing Deep Learning for Visual Perception

Related papers