itinai content

←back to Blog

The Roller Coaster SLAM Dataset: High-Dynamic Visual-Inertial Benchmarks from Amusement Rides

This is the world’s first SLAM dataset recorded onboard real roller coasters, offering extreme motion dynamics, perceptual challenges, and unique conditions for benchmarking SLAM algorithms under aggressive real-world trajectories.

Key Highlights:

Unprecedented Motion Dynamics – Captures high-acceleration motion with rapid velocity changes, sharp turns, and steep vertical drops, providing a stress test for visual-inertial odometry and SLAM systems.

Ground-Truth-Aided Evaluation – Includes precise track layouts and synchronized vehicle timings for accurate pose validation, supporting quantitative benchmarking under controlled, repeatable conditions.

Multisensor Payload – Features tightly time-synced IMU, monocular/stereo cameras, and optional GPS, enabling advanced sensor fusion research in high-dynamic, low-feature environments.

Perception Under Real-World Stress – Environments include tunnels, rapid lighting changes, and dynamic occlusions (e.g., riders, motion blur), pushing SLAM systems to their robustness limits.

Loop Closures & Structural Repetition – Tracks often revisit similar locations at different orientations and speeds, ideal for evaluating loop closure detection and map consistency.

Cross-Domain Applicability – Insights and techniques developed here transfer to drone flight, autonomous driving, and AR/VR in dynamic or aggressive motion scenarios.

Included Sequences:

  • TRON Lightcycle Power Run – Fast, low-light indoor/outdoor transitions with sharp direction changes.
  • Seven Dwarfs Mine Train – Moderate-speed sequence with rich structural textures and dynamic ride motion.

Data Sequence Info
Name: tron_2023-12-06
Format: .bag (ROS)
Download:

Device: OAK-D Pro W

  • Stereo: 640×400 @ 30 FPS
  • IMU: BMI270 @ 100 Hz
  • Timestamp offset: ~0.01 ms
  • Same boarding/alighting point: No
  • Intrinsics/Extrinsics: see camera_info, tf_static

IMU Allan Variance:

  • Accelerometer noise: 0.01 m/s²/√Hz | Random walk: 0.001 m/s³/√Hz
  • Gyroscope noise: 0.001 rad/s/√Hz | Random walk: 0.0001 rad/s²/√Hz

How to Collect & Contribute

To meet SLAM-quality standards, your dataset should include:

  • Calibrated intrinsics and extrinsics
  • Time-synced camera and IMU (preferably global shutter)
  • ROS bag export with proper metadata
  • GoPro (with IMU/GPS) – use tools like gopro_ros
  • OAK-D / Stereo Cameras – factory-calibrated, compact

Note: Most devices need external power and storage (e.g., UMPC or smartphone).

Mounting Tips:

  • Use secure mounts (head, chest, wrist)
  • Avoid handheld setups
  • Check ride vehicle shape beforehand if possible

Preferred Format:

  • ROS bags following REP standards (REP-104, REP-105, REP-145)
  • Include all calibration and transform info via CameraInfo and tf

Project Video: https://youtu.be/g6IYMR6LCec?feature=shared
Github: https://github.com/Factor-Robotics/Roller-Coaster-SLAM-Dataset

    1. Visual SLAM: https://learnopencv.com/monocular-slam-in-python/
    2. LiDAR SLAM: https://learnopencv.com/lidar-slam-with-ros2/
    3. MASt3R SLAM: https://learnopencv.com/mast3r-slam-realtime-dense-slam-explained/

The post The Roller Coaster SLAM Dataset: High-Dynamic Visual-Inertial Benchmarks from Amusement Rides appeared first on OpenCV.