The SFU Mountain Dataset: Semi-Structured Woodland Trails Under Changing Environmental Conditions

We present a novel long-term dataset of semistructured woodland terrain under varying lighting and weather conditions and with changing vegetation, infrastructure, and pedestrian traffic. This dataset is intended to aid the development of field robotics algorithms for long-term deployment in challenging outdoor environments. It includes more than 8 hours of trail navigation, with more available in the future as the environment changes. The data consist of readings from calibrated and synchronized sensors operating at 5Hz to 50Hz in the form of color stereo and grayscale monocular camera images, vertical and push-broom laser scans, GPS locations, wheel odometry, inertial measurements, and barometric pressure values. Each traversal covers approximately 4km across three diverse woodland trail environments, and we have recorded under four different lighting and weather conditions to date: dry; wet; dusk; night. We also provide 383 hand-matched location correspondences between traversals as ground-truth for benchmarking place recognition and mapping algorithms. This paper describes the configuration of the vehicle, the trail environments covered, and the format of the data we provide.