------------------- GENERAL INFORMATION ------------------- 1. Title of Dataset: The Canyonlands Dataset: Multi-Seasonal, Multimodal Field-Robotics Data for Ecological Monitoring of Degraded Rangelands 2. Abstract: A large-scale, multi-seasonal, multimodal field-robotics dataset collected with a teleoperated Clearpath Husky ("Restorebot") at the Canyonlands Research Center near Monticello, southeastern Utah. The same eighteen 10 m x 10 m experimental plots (six each of control, drill-seeded, and connectivity-modifier/"ConMod" treatments) were revisited in two seasons (May 2022, November 2022). Each plot provides forward- and downward-facing RGB imagery, 3D LiDAR point clouds, RTK-GPS, IMU, LIO-SAM odometry, sensor calibration, and per-frame GPS/compass-heading uncertainty, plus the raw ROS1 recordings. The data target long-term localization/mapping and cross-seasonal data association in unstructured, visually redundant, feature-sparse rangelands. 3. Authors: - Kristen Such — data collection, processing, lead curator — Dept. of Computer Science, University of Colorado Boulder - Doncey Albin — data collection, curation, hosting — Dept. of Computer Science, University of Colorado Boulder - Christoffer Heckman — Principal Investigator (corresponding) — Dept. of Computer Science, University of Colorado Boulder — ORCID 0000-0002-9651-6866 4. Contact information: christoffer.heckman@colorado.edu 5. Date of data collection: May 2022 and November 2022. 6. Associated grants/funding: - U.S. Department of Agriculture, National Institute of Food and Agriculture (USDA-NIFA), award COLW-2020-08988 (awarded through the National Robotics Initiative), PI C. Heckman. -------------------------- SHARING/ACCESS INFORMATION -------------------------- 1. Licenses/restrictions placed on the data: Creative Commons Attribution 4.0 International (CC BY 4.0). You are free to share (copy and redistribute) and adapt (remix, transform, build upon) this dataset for any purpose, including commercial, provided you give appropriate credit using the recommended citation above, link to the license, and indicate if changes were made. Full text: https://creativecommons.org/licenses/by/4.0/legalcode 2. Related publications and resources: - Such, K., Albin, D., Heckman, C. "RestoreBot: Two Years of Autonomous Field Robotics for Ecological Restoration of Degraded Rangelands." Under submission to IEEE Transactions on Field Robotics, 2026. (No DOI yet; this dataset record is a prerequisite for that submission.) - Toolkit & analysis code: https://github.com/arpg/canyonlands-dataset - Project page: https://arpg.colorado.edu/canyonlands_dataset/ 4. Recommended citation for the data: Heckman, C., Albin, D., & Such, K. (2026). Multi-Seasonal, Multimodal Field-Robotics Data for Ecological Monitoring of Degraded Rangelands [Dataset]. University of Colorado Boulder. https://doi.org/10.25810/AB1G-RK08 --------------------- DATA & FILE OVERVIEW --------------------- 1. File inventory & formats Total size: ~870 GB. On the order of ~780,000 files: - 446,306 RGB images (.png, compressed; filename = Unix timestamp in nanoseconds) - 167,071 LiDAR point clouds (.bin; float32 x, y, z, intensity; full 64x1024 Ouster scan) - ~167,000 per-frame GPS/odometry covariance matrices (.csv, 2x2, meters^2) - per-plot odometry, GPS/NMEA, IMU, frustum-corner, and compass-heading tables (.csv) - 36 raw ROS1 recordings (one per plot x season; .bag.lz4) - sensor calibration: URDF (.xacro) for each season; LIO-SAM config (.yaml); global maps (.bin) 2. Folder/directory structure Folder/directory structure: Canyonlands/ calib/ configs/ (LIO-SAM params .yaml; May/Nov URDF .xacro) May2022// (plot = {1..6}{conmod|control|drill}) Down_Facing_Images/ Front_Facing_Images/ PointClouds/ imu.csv nmea_sentences.csv processed_results/ (frustum_corners.csv, liosam_odometry.csv, global_map.bin, processed_compass_heading.csv, covariance_matrices/.csv) Nov2022// (Nov has Left_/Right_Down_Facing_Images instead of one) bags/ (raw ROS1 .bag.lz4, one per plot x season) 3. Software-specific information: - ROS1 (Noetic) for raw bag playback; LIO-SAM for odometry/mapping reproduction. - Python toolkit at the repo above (opencv, shapely, scipy, geographiclib, numpy, matplotlib); Docker for the included ROS1->ROS2 conversion. Raw components are also usable without ROS. -------------------------- METHODOLOGICAL INFORMATION -------------------------- 1. Research Methods: Methods / collection notes - Site: former cattle pasture at the Canyonlands Research Center (Redd family Dugout Ranch, managed by The Nature Conservancy), Monticello, UT. 18 plots, 10 m x 10 m; three corners of each surveyed (Trimble SPS855/Zephyr) and marked with rebar for cross-seasonal re-identification. - Procedure: the robot was teleoperated in a lawnmower pattern over each plot for full downward- camera coverage; campaigns timed to ecologically significant stages (May = early germination, November = seed stage). - Processing: LIO-SAM lidar-inertial odometry/mapping; GPS NMEA sentences republished to NavSatFix; the platform has no magnetometer, so compass heading was inferred from GPS course-over-ground plus hand-annotated forward-camera landmarks and projected with the Vincenty formula; per-frame GPS and heading uncertainty is provided as covariance. - Units / frames: positions in meters (odometry "map" frame; navigation x-forward, y-left, z-up); GPS in latitude/longitude (deg); timestamps are Unix nanoseconds. - Caveats: per-frame localization uncertainty has a median ~0.8 m (95th pct ~1.75 m) and compass-heading uncertainty a median ~90 deg; downward cameras sit ~13.5 cm above ground (~14 x 10 cm FoV) with no artificial illumination (shadows, motion blur). All 18 plots are present in both seasons; cross-seasonal trajectory overlap varies, with ~11 plots having sufficient overlap for frame-level cross-seasonal matching. Semantic/segmentation labels are NOT part of this data release. 2. Facilities or instruments: Instruments / sensors (2022 "Restorebot" platform, Clearpath Husky A200): - Ouster OS1-64 LiDAR (64 beams, 360 deg x ~33 deg FoV, up to 100 m, 20 Hz) - Intel RealSense D435 cameras (forward + downward; 2x downward in Nov 2022) - LORD MicroStrain 3DM-GX5-15 IMU (VRU) - Trimble BX922 GPS + AG25 antenna, 5 cm RTK/RTX (CenterPoint RTX), ~20 Hz - Trimble SPS855 base + Zephyr Model 3 antenna (survey-grade plot-corner ground truth) - Onboard compute: AMD Ryzen Threadripper 3990X + NVIDIA GTX 1650