CRADMap: Applied Distributed Volumetric Mapping with 5G-Connected Multi-Robots and 4D Radar Perception

TL;DR

CRADMap tackles the challenge of acquiring dense, globally consistent 3D maps in multi-robot indoor scenarios with limited onboard compute. It fuses an ORBSLAM3-based front-end on each AMR with a centralized COVINS backend that performs global pose optimization, while streaming dense keyframes over 5G to enable real-time volumetric reconstruction across robots. A standalone 4D mmWave radar BtV module provides occluded-object detection, generating an independent radar map to augment scene understanding without fusion into RGB-D data. Experimental results on UW-E7 and TUM RGB-D benchmarks show substantial gains in environmental coverage (roughly 78.9%–85%) and point density (4.8×–5.5×) over sparse baselines, with 5G providing reliable bandwidth and low latency; BtV radar achieves robust detection of metallic structures behind occlusions (84%–92%). The work demonstrates a practical, scalable pathway for high-detail, real-time multi-robot mapping with enhanced BtV perception for inspection and asset-management tasks.

Abstract

Sparse and feature SLAM methods provide robust camera pose estimation. However, they often fail to capture the level of detail required for inspection and scene awareness tasks. Conversely, dense SLAM approaches generate richer scene reconstructions but impose a prohibitive computational load to create 3D maps. We present a novel distributed volumetric mapping framework designated as CRADMap that addresses these issues by extending the state-of-the-art (SOTA) ORBSLAM3 system with the COVINS on the backend for global optimization. Our pipeline for volumetric reconstruction fuses dense keyframes at a centralized server via 5G connectivity, aggregating geometry, and occupancy information from multiple autonomous mobile robots (AMRs) without overtaxing onboard resources. This enables each AMR to independently perform mapping while the backend constructs high-fidelity real-time 3D maps. To operate Beyond the Visible (BtV) and overcome the limitations of standard visual sensors, we automated a standalone 4D mmWave radar module that functions independently without sensor fusion with SLAM. The BtV system enables the detection and mapping of occluded metallic objects in cluttered environments, enhancing situational awareness in inspection scenarios. Experimental validation in Section~\ref{sec:IV} demonstrates the effectiveness of our framework.

Figures (5)

• Figure 1: Overview of our approach. The left section shows the data transmission cycle for map visualization. The center highlights an AMR turtlebot4 with its key integrated components. The right section presents a 360-degree top-view CRADMap of the UW RoboHub lab, while the bottom right shows the radar point cloud map detecting an obscured vent pipe.
• Figure 2: Comprehensive pipeline flow of the proposed methodology in three modules. The gray module illustrates the structure of the AMRs nodes. The blue module represents the frontend and backend stages of the distributed framework for multi-robot volumetric maps and radar point cloud generation, which are managed on a central server to offload computational processing. The black layer in the middle highlights the real-time data transmission stage.
• Figure 3: Qualitative comparison of our dense volumetric mapping approach (top row) with the baseline ORB-SLAM3 campos2021orb sparse map (middle row). The bottom row shows the field of view and ORB feature detection by the SLAM frontend. The dataset was captured live on different floors of the UW Engineering (UW-E7) building for evaluation.
• Figure 4: Cluttered indoor setting with furniture and vent pipe present horizontal with the floor. Only 4D mmWave radar detects successfully in (b).
• Figure 5: In a completely blocked view, 4D mmWave radar sensing shows a point cloud map of three metal studs behind the handcrafted wall in (c).