A multi-layer and multi-sensor Localization and Mapping approach for agricultural environments.
VineSLAM is a localization and mapping algorithm designed for challenging agricultural environments such as mountain vineyards and flat orchards. This algorithm is based on two main performance optimizers: (1) topological mapping to manage the memory resources and enable large-scale and long-term operation; and (2) CUDA-based GPU optimizations to improve runtime performance.
This algorithm supports two operation modes: SLAM, where a multi-layer map is built while the robot is simultaneously localized; and localization-only, using a pre-built map to localize the robot.
This implementation is currently supported in ROS2 Foxy and ROS2 Eloquent.
- Version History
- Prototype Status
- Software Roadmap
- Dependencies
- Installation Instructions
- Usage
- Project Structure
- Contributors
- Acknowledgments
- License
Below is a list of versions released for this project. Each version has a link to its release notes or relevant documentation.
| Version | Release Date | Description |
|---|---|---|
| v1.0.0 | 2024-11-11 | • Fully compatible with ROS2 foxy • Sensors supported: IMU, GNSS, 3D LiDAR, 2D Odometry • Validation in ROSBAG2 datasets: fully tested and validated in indoor/outdoor environments • Deployment on robots in real time: fully tested and validated in indoor environments |
- In Development: aligned with Roadmap.
- In Testing Phase: being fully tested in indoor/outdoor urban environments.
The planned milestones and tasks for the project's development. Completed tasks are marked.
- ROS2 Compatibility
- Compatible with ROS2 foxy
- Compatible with ROS2 humble
- Fully working on indoor/outdoor 3D urban environmets
- Precise mapping in indoor/outdoor 3D urban environments
- Precise 3D localization on the robot in real-time in indoor/outdoor 3D urban environments - extensive testing
- Fully working on outdoor 3D complex agricultural environments
- Precise mapping in outdoor 3D complex agricultural environments
- Short-term tests (<500m) on the robot in real-time in outdoor 3D complex agricultural environments
- Long-term tests (>500m) on the robot in real-time in outdoor 3D complex agricultural environments
- Particle Filter (PF) improvement
- Correct model state to support pure rotations misaligned with robot's centre of mass
- Improve PF performance on harsh inclinations (roll and pitch)
- Semantic Data
- Create mapping process for semantic data in agriculture: trunks, fruits, leaves, and others
- Create filtering process for dynamic data: persons, animals, cars, and others
- Use semantic mapping process in the localization loop
- Integrate Loop Closure
- Explore open-source Loop Closure frameworks
- Integrate Loop Closure on VineSLAM using the Topological Map concept
- Validate improvements of the mapping process with this implementation
- Software Dependencies: OpenCV, Boost, Asio, ros-foxy-full, ros-foxy-diagnostic-updater, ros-foxy-vision-msgs, ros-foxy-ublox-msgs
To install them, run the following commands:
# General dependencies
sudo apt-get install libopencv-dev
sudo apt-get install libboost-dev
sudo apt-get install libasio-dev
# ROS dependencies (considering a ros-foxy-full installation)
sudo apt-get install ros-foxy-diagnostic-updater
sudo apt-get install ros-foxy-vision-msgs
sudo apt-get install ros-foxy-ublox-msgs- Optional Dependencies: To use the GPU optimized version, you must install CUDA, from https://developer.nvidia.com/cuda-downloads.
Follow these steps to set up the project in your local environment.
- Clone the repository:
cd <path_to_ros2_ws>/src git clone https://gitlab.inesctec.pt/agrob/vineslam_stack/vineslam -b master
- Compile VineSLAM:
cd <path_to_ros2_ws> colcon build --cmake-args -DCMAKE_BUILD_TYPE=Release -DLIDAR_TYPE=0
The -DLIDAR_TYPE cmake argument compiles VineSLAM for specific LiDAR models:
-DLIDAR_TYPE=0corresponds to Velodyne VLP16-DLIDAR_TYPE=1corresponds to RoboSense Helios 5515-DLIDAR_TYPE=2corresponds to Livox Mid 70-DLIDAR_TYPE=3corresponds to Ouster-OS-1-64-DLIDAR_TYPE=4uses all the features of the subscribed point cloud- The default value is 0, if no
-DLIDAR_TYPEis chosen
Note that if you have CUDA installed, VineSLAM will automatically build CUDA kernels and use GPU optimizations.
To install and build VineSLAM you can also use the provided Dockerfile. Note that the Docker environment does not consider CUDA installation.
git clone https://gitlab.inesctec.pt/agrob/vineslam_stack/vineslam -b master
cd vineslam/docker
docker build -t vineslam .This will create a docker container with VineSLAM ready to compile and execute. Now you just have to...
docker run -it vineslam bash
colcon build --cmake-args -DCMAKE_BUILD_TYPE=ReleaseBefore executing VineSLAM, please check the Interfaces file. In this you will find the published/subscribed topics, the /tf listened and broadcasted, and the parameters and their meaning. With this, you can better tune and use VineSLAM.
After performing the topic remappings and parameters tuning you can simply execute, for example the SLAM Node, as follows:
cd <path_to_ros2_ws>/src/vineslam/test/test_slam_node
ros2 launch run.launch.pyThis will automatically open a rviz visualization panel properly configured.
The same can be done for the localization node in the respective test folder.
| Folder/File | Description |
|---|---|
/docker |
Contains the docker files needed to build vineslam in a container |
/docs |
Contains the images used in this README file and the ROS2 interfaces used by VineSLAM |
/test |
Contains two test folders to easily execute VineSLAM in SLAM and Localization modes and an utilities folder that contains debug nodes |
/vineslam |
Implements all the VineSLAM algorithms (mapping, localization, feature extraction, and others) |
/vineslam_ros |
Implements all the VineSLAM ROS2 interfaces (publishers, subscribers, services) |
We thank the following team members for their contributions to this project:
| Name | |
|---|---|
| André Silva Aguiar | andre.s.aguiar@inesctec.pt |
| Filipe Neves dos Santos | fbsantos@inesctec.pt |
This prototype was funded and developed under the following projects:
- SCORPION - SCORPION’s objective is to develop a safe and autonomous precision spraying tool integrated into a modular unmanned tractor (robotics platform) to increase spraying efficiency, while reducing human and animal exposure to pesticides, water usage and labour costs.
- ROMOVI - The RoMoVi project main objective is the development of robotic components and a modular and extensible mobile platform, which will allow in the future to provide commercial solutions for hillside vineyards capable of autonomously executing operations of monitoring and logistics.
- ROSIN - ROS-Industrial aims to transfer value and the ease of application to industrial hardware, by developing new components, improving existing ones, but also by performing non-development work such as compiling usage and development guidelines or performing license audits.
This project is licensed under the GNU General Public License v3.0 - see the LICENSE file for details.