Roller Coaster SLAM Dataset

This project aims to collect data on roller coasters Around The World to make SLAM Harder Better Faster Stronger.

Why are roller coasters an ideal platform for SLAM research?

Generated by ChatGPT:

Roller coasters can be considered as an ideal platform for SLAM (Simultaneous Localization and Mapping) research due to several reasons:

1. Controlled Environment: Roller coasters operate within a controlled environment, where the track layout is predetermined and the position of the vehicle is precisely known at all times. This allows researchers to collect accurate data and validate their SLAM algorithms in a controlled setting.

2. Dynamic Motion: Roller coasters involve fast-paced, dynamic motion with rapid changes in speed, acceleration, and direction. This challenges SLAM algorithms to accurately estimate the vehicle's pose and map the surroundings while dealing with significant motion disturbances.

3. Sensor Fusion: Roller coasters provide opportunities for sensor fusion by integrating multiple sensors such as GPS, IMU (Inertial Measurement Unit), LIDAR (Light Detection and Ranging), cameras, etc., on the coaster cars or along the track. Sensor fusion enables better localization accuracy and richer mapping information.

4. Perception Challenges: Roller coasters often traverse through various terrains, tunnels, obstacles, and structures. These diverse environments present perceptual challenges for SLAM algorithms to handle occlusions, loop closures, feature extraction, mapping dynamic objects (such as other riders or operators), etc.

5. Real-Time Applications: Roller coaster SLAM research can have practical applications beyond amusement parks. It can be applied to autonomous vehicles navigating complex environments where high-speed maneuvers are required while maintaining accurate localization and mapping capabilities.

By utilizing roller coasters as a testbed for SLAM research, scientists and engineers can develop more robust algorithms that improve navigation systems in various real-world scenarios beyond amusement parks.

How to collect data and contribute?

The difference between SLAM data sequences and many roller coaster POV videos on the Internet lies in the following aspects:

1. Sensor intrinsics and extrinsics need to be provided.
2. IMU data is required in most cases.
3. Cameras typically use a global shutter.
4. Data from different sensors needs to be synchronized.

The instructions below are mainly about camera devices, as they are generally allowed in amusement parks. We currently have no experience or cases of installing LiDAR on roller coasters. Perhaps some small 3D reconstruction device equipped with LiDAR is feasible. If you can get advanced access to the roller coaster, then everything is possible.

Available Devices

GoPro and other action cameras

New GoPro models all come with IMU, and some also have GPS. Its data can be converted into a ROS bag using tools such as gopro_ros. Other tools may be useful for sensor calibration, such as OpenICC.

OAK and other stereo cameras

Many stereo cameras used in robots are also small and have intergrated IMUs. Most of them are already calibrated at the factory. But they usually don't have built-in batteries and storage. Therefore you need other devices for power supply and data recording, such as a mobile phone or UMPC (Ultra-Mobile Personal Computer). The most important thing is to organize the cable and keep yourself and your devices safe during the ride.

Mounting Accessories

Holding the device directly with your hands is not recommended. You need some accessories to mount the device securely. You can find a variety of action camera accessories for different mounting methods. Commonly considered mounting positions include head, chest, and wrist, depending on the shape of the vehicle and restrains. Some amusement parks will display the vehicle next to the project. This is helpful because you can comfirm in advance whether the mounting method is suitable.

Data Format

It is recommended to share data in the form of ROS bags. Because it can record time series data from various types of sensors and has good specifications. Sensor intrinsics and extrinsics can be recorded in tf and sensor_msgs/CameraInfo without extra provision. We recommend following relevant REPs (ROS Enhancement Proposals) to avoid causing trouble to users, including:

• REP-104 CameraInfo updates for Diamondback
• REP-105 Coordinate Frames for Mobile Platforms
• REP-117 Informational Distance Measurements
• REP-118 Depth Images
• REP-145 Conventions for IMU Sensor Drivers

It is best to also attach the project’s official webpage or introduction, if available. There will generally be information such as manufacturer, height, length, maximum speed, etc., which can also provide a useful reference. When this project collects more sequences, we will consider organizing them according to geographical location. So that everyone can find local projects for more diverse records and repeated tests.