Sunday 02 February 2025
A team of researchers has created a new dataset called IRisPath, which is designed to help autonomous vehicles navigate off-road terrain more effectively. The dataset consists of images taken by cameras and LiDAR sensors mounted on a vehicle as it drives through various environments during both day and night.
The dataset is particularly useful because it includes both infrared (IR) and RGB images, which are able to capture different types of information about the environment. IR images are more resistant to changes in light and weather conditions, making them better suited for use at night or in foggy or dusty conditions. RGB images, on the other hand, provide rich semantic information about the terrain.
The researchers used a novel method to calibrate the extrinsic parameters of the cameras and LiDAR sensors, allowing them to register the IR and RGB images with the 3D point cloud data from the LiDAR sensor. This process is typically challenging because it requires identifying common features between the two modalities, but the researchers developed a targetless method that uses only the camera motion estimation.
The dataset was collected using a vehicle equipped with multiple sensors, including FLIR’s Hadron Camera and Livox’s Mid360 LiDAR. The researchers used this data to train a model that can predict the traversability of different terrain types, taking into account factors such as speed and weather conditions.
The model uses a fusion approach that combines the IR and RGB images with the vehicle’s state feedback to generate a costmap, which represents the difficulty of navigating through a particular area. The researchers found that this approach improved the accuracy of the costmap compared to using only single-modality data.
To evaluate the performance of the model, the researchers used synthetic weather augmentations and tested it on different types of terrain. They also analyzed various state-of-the-art test time adaptation methods and found that their proposed method outperformed them in terms of costmap accuracy.
The IRisPath dataset is an important step towards developing more advanced autonomous vehicles that can navigate complex off-road environments. The researchers hope that their dataset will be useful for other researchers and developers working on similar projects, and they plan to release it publicly to facilitate further research and development.
Cite this article: “IRisPath: A Novel Dataset for Off-Road Autonomous Vehicle Navigation”, The Science Archive, 2025.
Autonomous Vehicles, Off-Road Terrain, Irispath Dataset, Cameras, Lidar Sensors, Infrared Images, Rgb Images, Extrinsic Calibration, Targetless Method, Costmap Generation
