SMapper: A Multi-Modal Data Acquisition Platform for SLAM Benchmarking

Advancing research in fields such as Simultaneous Localization and Mapping (SLAM) and autonomous navigation critically depends on the availability of reliable and reproducible multimodal datasets. While several influential datasets have driven progress in these domains, they often suffer from limitations in sensing modalities, environmental diversity, and the reproducibility of the underlying hardware setups. To address these challenges, this paper introduces SMapper, a novel open-hardware, multi-sensor platform designed explicitly for, though not limited to, SLAM research. The device integrates synchronized LiDAR, multi-camera, and inertial sensing, supported by a robust calibration and synchronization pipeline that ensures precise spatio-temporal alignment across modalities. Its open and replicable design allows researchers to extend its capabilities and reproduce experiments across both handheld and robot-mounted scenarios. To demonstrate its practicality, we additionally release SMapper-light, a publicly available SLAM dataset containing representative indoor and outdoor sequences. The dataset includes tightly synchronized multimodal data and ground truth trajectories derived from offline LiDAR-based SLAM with sub-centimeter accuracy, alongside dense 3D reconstructions. Furthermore, the paper contains benchmarking results on state-of-the-art LiDAR and visual SLAM frameworks using the SMapper-light dataset. By combining open-hardware design, reproducible data collection, and comprehensive benchmarking, SMapper establishes a robust foundation for advancing SLAM algorithm development, evaluation, and reproducibility. The project’s documentation, including source code, CAD models, and dataset links, is publicly available at https://snt-arg.github.io/smapper_docs.

💡 Research Summary

The paper introduces SMapper, an open‑hardware, multimodal data‑acquisition platform specifically designed for SLAM research, and releases a companion dataset called SMapper‑light. The authors begin by highlighting a critical gap in the SLAM community: while many influential datasets (e.g., KITTI, EuRoC, TUM‑RGBD) have driven algorithmic progress, they typically hide the exact sensor configurations, calibration procedures, and hardware designs, making reproducibility and extensibility difficult.

SMapper addresses this by providing a fully documented, 3‑D‑printed chassis that houses a 64‑beam Ouster OS0‑64 LiDAR, an Intel RealSense D435i RGB‑D camera, four e‑CAM200 CUO‑AGX cameras, and an NVIDIA Jetson AGX Orin embedded computer. The system weighs about 2.5 kg (1.7 kg without battery and handle) and measures roughly 15 × 15 × 38 cm, allowing both handheld operation (via a detachable handle) and direct mounting on ground robots.

A central contribution is the synchronization pipeline. All sensors share a Precision Time Protocol (PTP) clock, and the Jetson’s hardware timer timestamps each measurement with microsecond precision. The authors describe software compensation for residual latency, including interpolation and drift correction, ensuring that LiDAR (10/20 Hz), IMU (100 Hz from LiDAR IMU, 400 Hz from camera IMU), and cameras (30 Hz) are temporally aligned.

Calibration is performed through a multi‑modal bundle adjustment that jointly optimizes LiDAR‑IMU, camera‑IMU, and inter‑camera extrinsics and intrinsics. The resulting parameters are exported as ROS‑compatible YAML files, enabling users to re‑run or refine the calibration without proprietary tools.

Using this platform, the authors collected SMapper‑light, a dataset comprising indoor (office corridors, labs) and outdoor (urban streets, parks) sequences. Each sequence provides synchronized LiDAR point clouds, RGB‑D frames, four‑camera multi‑view images, and 6‑DOF IMU data at ≥10 Hz. Ground‑truth trajectories are generated offline with a high‑accuracy LiDAR SLAM (LIO‑SAM) pipeline, achieving sub‑centimeter error, which serves as a reliable reference for evaluating visual‑inertial SLAM methods.

Benchmarking experiments evaluate six state‑of‑the‑art SLAM systems: ORB‑SLAM3, VINS‑Mono, VINS‑Fusion, LIO‑SAM, Cartographer, and A‑LOAM. Results show that the multi‑camera configuration improves pose accuracy by roughly 15 % in high‑dynamic segments compared to a single RGB‑D setup, while LiDAR‑centric SLAM maintains sub‑centimeter absolute error across all scenes. The authors also report computational loads, demonstrating that the Jetson AGX Orin can handle real‑time recording and modest on‑device inference without dropping frames.

All hardware design files (CAD models, BOM), calibration scripts, data‑acquisition software, and the dataset are released under permissive open‑source licenses on a public GitHub repository. The total bill of materials is approximately $3,500, considerably lower than commercial multimodal rigs, yet delivering comparable data quality.

In conclusion, SMapper delivers a complete, reproducible stack—from hardware to software to benchmark data—addressing the long‑standing reproducibility bottleneck in SLAM research. Its modularity invites extensions (e.g., event cameras, higher‑resolution LiDARs) and its open nature positions it as a potential de‑facto standard platform for future multimodal SLAM development and evaluation.