FRUC multiple sensor forest dataset including absolute, map-referenced localization

FRUC Datasets (Forest environment dataset)

This dataset was collected as part of the work conducted by the Forestry Robotics @ University of Coimbra team (https://www.youtube.com/@forestryroboticsuc; part of the Institute of Systems and Robotics, https://www.isr.uc.pt/) within the scope of the Safety, Exploration and Maintenance of Forests with Ecological Robotics (SEMFIRE, ref. CENTRO-01-0247-FEDER-03269; http://semfire.ingeniarius.pt/) and the Semi-Autonomous Robotic System for Forest Cleaning and Fire Prevention (SafeForest, CENTRO-01-0247-FEDER-045931) research projects. Its purpose is to allow researchers in forestry robotics to have an in-depth analysis of a florests environment; obtain an a priori map for robot operations (e.g. path plannning, landscaping, etc…) and to train segmentation algorithms;

The dataset in question includes data from multiple sensors and absolute, map-referenced localization which can be used to register the sensor data to a fixed coordinate system. It was collected at the Choupal National Woods, Coimbra, Portugal (40^◦13′13.3′′N;8^◦26′38.1′′W). The dataset was collected during a partly clouded day in a forest environment by performing two circular loop laps amounting to a total distance of approximately 800m, with a total duration of 14 minutes and 22 seconds. The scenario is rich in features relevant to forestry robotics applications, including trees, bushes, tree trunks, etc. To better handle the multimodal nature of the acquired data, the dataset is bundled into rosbags, a file format used by the ROS (Robot Operating System) to record and play back data.

More specifically, the datasets include:

• RGB Images from an Intel Realsense D435i
• Aligned Depth Images from an Intel Realsense D435i
• Left and Right Mono Images from a Mynt Eye s1030
• Point Clouds from a Livox Mid-70 LiDAR
• Unfiltered acceleration, gyroscopic and magnetic data from a Xsens MTi IMU
• Unfiltered acceleration, gyroscopic data from an Intel Realsense D435i
• GNSS Fix data from a Xiaomi Mi Mix 3 device

Description of files:

1. The dataset is contain in choupal.bag.
2. The rosbag_info.txt contains the information of each rosbag;
3. The sensor_box.urdf contains all the required transforms;
4. The sensor_box.stl contains the 3D model of the apparatus;
5. The choupal.launch publishes the sensor transforms and plays the dataset;
6. The localization.bag contains the final graph of poses extracted with Cartographer republished as nav_msgs/odom at 4.98Hz.
7. The localization_15Hz.bag contains a map-referenced localization extracted with Cartographer at a higher frequency, but the poses are interpolated. If you don't require a high frame rate, please use the localization.bag instead.

Usage:

1. Extract the fruc_dataset_choupal_launch.zip into a catkin workspace
2. Install the necessary dependencies of the package:

   1. cd [/path/to/catkin_ws]

       

   2. rosdep install --from-paths src --ignore-src -y -r

3. Copy the rosbags into the fruc_dataset_choupal_launch/rosbag/
4. Edit the fruc_dataset_choupal_launch/launch/choupal.launch file to your use case:
   1. Change the file_path argument if the rosbags are not in the default location;
   2. Set localization_file to  the path of the desired localization bag, leave it empty to run the dataset without localization.
5. Compile the package and source the environment:

   1. catkin_make [/your_catkin_workspace/]

       

   2. source [/your_catkin_workspace/devel/setup.bash]

       

6. Launch the files:

   roslaunch fruc_dataset_choupal_launch choupal.launch