Hilti-Research/hilti-slam-challenge-2023

Hilti SLAM Challenge 2023

The Hilti SLAM Challenge 2023 dataset is a multimodal robotics benchmark for evaluating simultaneous localization and mapping (SLAM) and sensor-fusion systems in challenging, real-world construction environments.

The 2023 challenge extends the previous Hilti SLAM benchmarks to multi-session and multi-platform SLAM. Recordings were collected across multiple active construction sites, with overlapping trajectories captured during different sessions and using different sensor platforms.

The benchmark supports both:

• —single-session SLAM, where each trajectory is evaluated independently; and
• —multi-session SLAM, where information from multiple runs, and in some cases multiple sensor platforms, is combined into a consistent map of a construction site.

Two acquisition platforms are included: the handheld Phasma-style sensor platform used in the 2022 challenge and a new sensor suite mounted on a tracked drilling robot.

Questions and implementation discussions can be raised in the Hilti SLAM Challenge 2023 GitHub repository.

Photosensitivity warning: the additional handheld Site 2 sequences contain rapidly flashing lights in their video streams. These recordings may affect viewers who are sensitive to flashing imagery.

Dataset contents

The dataset contains 12 primary challenge sequences distributed across three construction sites, together with 3 additional handheld sequences for Site 2.

Each site forms a multi-session SLAM group with overlap between individual trajectories. At the same time, every trajectory can be processed and evaluated independently as a conventional single-session SLAM sequence.

Sensor platforms

Handheld sensor suite

The handheld platform contains:

• —a Sevensense Alphasense Core camera head with five 0.4 MP global-shutter cameras;
• —a Hesai PandarXT-32 LiDAR; and
• —the associated inertial sensing system.

The sensors are rigidly mounted on an aluminium handheld platform.

Camera synchronization is performed using an FPGA, while the cameras and LiDAR are synchronized using PTP. Sensor timing is aligned to within approximately 1 ms.

An external steel pin is attached to the platform for reference measurements.

Robot sensor suite

The robot-mounted sensor suite contains:

• —a RoboSense BPearl hemispherical LiDAR;
• —an Xsens MTi-670 IMU; and
• —four OAK-D cameras.

The sensors are mounted on a rigid frame installed on a tracked drilling robot platform.

Synchronization uses a combination of PTP and hardware triggering. The IMU and LiDAR clocks are aligned to within approximately 1 ms, while the IMU and camera clocks are aligned to within approximately 2 ms.

Ground truth

All challenge sequences are accompanied by 3-DoF reference measurements for trajectory evaluation.

Ground-truth files are provided separately from the ROS recordings. Users should inspect the supplied calibration and coordinate-frame definitions before comparing estimated trajectories with the reference measurements.

For multi-session experiments, trajectories belonging to the same construction site contain overlapping observations and can be jointly processed to aggregate map information across sessions.

Calibration

Calibration resources are provided for both acquisition platforms.

The official dataset distribution includes:

• —sensor calibration parameters;
• —a calibration sequence for the handheld platform;
• —a calibration sequence for the robot platform; and
• —the calibration-board YAML definition.

These recordings can be used to inspect or reproduce the supplied calibration.

Links

• —Official Hilti SLAM Challenge 2023 dataset page
• —Hilti SLAM Challenge 2023 GitHub repository
• —Hilti SLAM Challenge 2023 paper

Citation

When using this dataset in academic work, please cite:

@misc{nair2024hiltislamchallenge2023,
      title={Hilti SLAM Challenge 2023: Benchmarking Single + Multi-session SLAM across Sensor Constellations in Construction}, 
      author={Ashish Devadas Nair and Julien Kindle and Plamen Levchev and Davide Scaramuzza},
      journal={IEEE Robotics and Automation Letters},
      number={8},
      pages={7286–7293},
      year={2024},
      DOI={10.1109/lra.2024.3421791}     
}

License

The dataset is released under the Creative Commons Attribution-NonCommercial-ShareAlike 3.0 license (CC BY-NC-SA 3.0).

Use is limited to non-commercial purposes. Attribution is required, and adaptations or derivative works must be distributed under the same or a compatible license.

Consult the full license text and the official challenge page before redistribution or publication of derived data.