AI4Mars: A Dataset for Terrain-Aware Autonomous Driving on Mars

This dataset was built for training and validating terrain classification models for Mars, which may be useful in future autonomous rover efforts. It consists of ~425K semantic segmentation full image labels and ~50K images from Perseverance, Curiosity, Opportunity, and Spirit rovers, collected through crowdsourcing. Each image was labeled by ~5-10 people to ensure greater quality and agreement of the crowdsourced labels. It also includes ~1.5K validation labels annotated by the rover planners and scientists from NASA’s MSL (Mars Science Laboratory) mission, which operates the Curiosity rover, and MER (Mars Exploration Rovers) mission, which operated the Spirit and Opportunity rovers.

NOTE: the 10 per-subject labels included in the dataset have been combined, resulting in a total of ~50K actual labels provided. The unmerged dataset is provided for context, but is uncleaned and may contain some invalid data. Labels will visually look like black and white images: semantic classes have pixel values of 0-3. Further information on dataset usage may be found in the related papers and in the documentation files contained in the dataset.

Data was collected using the Zooniverse crowdsourcing platform.

Attribution:
If using this dataset in your work, please cite one of the related papers.