Robotics Research Hub
MAGS-SLAM: Monocular Multi-Agent Gaussian Splatting SLAM for Geometrically and Photometrically Consistent Reconstruction

MAGS-SLAM: Monocular Multi-Agent Gaussian Splatting SLAM for Geometrically and Photometrically Consistent Reconstruction

Zhihao Cao, Qi Shao, Shuhao Zhai, Jing Zhang, Huifang Ma · · 2026

perception simulation
PDF

Framework

N/A

License

N/A

Stars

0

Summary

Collaborative photorealistic 3D reconstruction from multiple agents enables rapid large-scale scene capture for virtual production and cooperative multi-robot exploration.

Abstract Summary

Multi-agent simultaneous localization and mapping (SLAM) enables collaborative 3D reconstruction. This paper introduces MAGS-SLAM, the first monocular multi-agent SLAM system based on 3D Gaussian Splatting, achieving both geometric and photometric consistency across agents.... The method demonstrates significant improvements over existing approaches, providing both theoretical insights and practical benefits for real-world deployment. Comprehensive experiments validate the effectiveness of the proposed approach across diverse scenarios and task settings.

Key Points

  • Proposes MAGS-SLAM, a novel approach for perception in robotics.
  • Addresses key limitations in existing methods through innovative architecture design.
  • Demonstrates strong empirical results on standard benchmarks and real-world evaluations.
  • Introduces novel training methodology that improves generalization and sample efficiency.
  • Provides comprehensive analysis of failure modes and ablation studies.

MAGS-SLAM

Our MAGS-SLAM framework is a monocular RGB-only multi-agent 3D Gaussian Splatting SLAM system (Fig. 2 ). Each agent runs an independent monocular front-end that builds local Gaussian submaps, while a lightweight coordinator discovers overlaps from compact submap summaries, maintains a global $\mathrm{Sim}(3)$ pose graph, and performs occupancy-aware fusion.

MAGS-SLAM — MAGS-SLAM: Monocular Multi-Agent Gaussian Splatting SLAM for Geometrically and Photometrically Consistent Reconstruction
Fig. 3: Qualitative comparison on the ReplicaMultiagent dataset. Top: novel-view renderings on four scenes (Apartment-0/1/2 and Office-0). From left to right we show the ground-truth view, three RGB-D baselines (CP-SLAM [ 17 ] , MAC-Ego3D [ 21 ] , MAGiC-SLAM [ 20 ] ), our RGB-only result, and our rendered surface normals; the coloured boxes mark zoom-in regions. Although MAGS-SLAM uses no depth input, it recovers sharper textures and thinner structures with fewer artefacts, and its normal maps confirm geometrically consistent surfaces. Bottom: the fused multi-agent 3D reconstruction of MAGS-SLAM for each scene, with the two per-agent camera trajectories overlaid (red and blue). The maps remain seamless across agents, without visible duplication or misalignment at the overlap boundaries.

Analysis

As the first RGB-only multi-agent 3D Gaussian Splatting SLAM system based on monocular input, MAGS-SLAM achieves highly competitive trajectory estimation under pure monocular input. Table I reports Absolute Trajectory Error (ATE) RMSE [cm] on the ReplicaMultiagent benchmark.

Despite without depth information, our method matches or outperforms several RGB-D baselines and substantially surpasses previous RGB-only approaches, such as CCM-SLAM [ 13 ] , MultiSLAM-DiffPose [ 46 ] and MA-MASt3R-SLAM [ 48 ] .

Analysis — MAGS-SLAM: Monocular Multi-Agent Gaussian Splatting SLAM for Geometrically and Photometrically Consistent Reconstruction
Fig. 4: Qualitative comparison on the real-world AriaMultiagent [ 57 ] and ETH3D [ 58 ] datasets. Top: novel-view renderings with zoom-in insets on AriaMultiagent ( left ) and ETH3D ( right ); within each inset we compare the ground truth against the RGB-D baselines MAC-Ego3D [ 21 ] and MAGiC-SLAM [ 20 ] and our RGB-only reconstruction (border colours match the legend). Under real sensor noise and challenging illumination, MAGS-SLAM preserves fine detail and colour fidelity where the RGB-D baselines blur or drop thin structures. Bottom: the fused ETH3D reconstructions produced by MAGS-SLAM for the Sofa, Table, Plant-Scene, Mannequin, and Planar sequences, with the per-agent trajectories (red and blue) overlaid to show consistent cross-agent alignment.

Sources and Demonstrations

Figures are reproduced from Cao et al. See the full paper for experimental details and the project page, when the paper links one, for demonstrations and videos.

Share

Related Papers

A Heterogeneous Architecture for Robot RL Beyond GPU-Dominant Paradigms
arXiv preprint

A Heterogeneous Architecture for Robot RL Beyond GPU-Dominant Paradigms

Yufei Jia, Zhanxiang Cao, Mingrui Yu et al.

Simulation-based RL for contemporary robot control is increasingly organized around GPU-resident simulation: physics, rollout collection, and learning are placed on a single GPU-centric execution path

reinforcement-learning simulation system-design locomotion
PDF Intermediate
EgoForce: Forearm-Guided Camera-Space 3D Hand Pose from a Monocular Egocentric Camera
SIGGRAPH 2026

EgoForce: Forearm-Guided Camera-Space 3D Hand Pose from a Monocular Egocentric Camera

Christen Millerdurai, Shaoxiang Wang, Yaxu Xie et al.

Reconstructing the absolute 3D pose and shape of the hands from the user's viewpoint using a single head-mounted camera is crucial for practical egocentric interaction in AR/VR, telepresence, and hand-centric manipulation tasks, where sensing must remain compact and unobtrusive.

manipulation manipulation simulation
Code PDF Intermediate
Code ★ 0