The need for robotics in agriculture is becoming more and more evident. Not only would robotic systems be able to take over menial, repetitive, and arduous tasks, but they would also be able to boost productivity. One area in this field that focuses specifically on collecting data to make informed management decisions regarding efficiency, productivity, quality, and profitability of crops is precision agriculture. But, there is only so much that humans can do on their own to gather the data required to drive this, and those processes can be extremely time-consuming and open to human error.
While the development of an autonomous navigation system for the implementation of robots in agriculture applications is by no means novel, many of these state-of-the-art solutions require expensive sensors such as high-grade 3D LiDARs or GNSS receivers. Such a high-cost barrier to entry is stifling robotics growth in the agriculture industry and is gating such developments for smaller profit farmers or businesses.
Low-Price Yet High Quality
Recognizing this gap, the PIC4SeR Centre from the Polytechnic University of Turin is using Jackal UGV and Husky UGV to drive their research towards developing a reliable and low-cost autonomous navigation system for row-based crops (e.g. orchards and vineyards). In particular, their solution exploits a low-cost minimum set of sensors: a GNSS receiver, an IMU, and an RGB-D camera, along with deterministic and artificial intelligence algorithms, to achieve robust and reliable autonomous navigation in vineyards and orchards. A large focus of their research is to achieve the same level of robustness and reliability as standard autonomous navigation systems while utilizing low-cost and off-the-shelf sensors.
“The main motivation is the ROS-support that allows us to easily deploy ROS- compatible code without taking care of low-level communication among different software packages. Moreover, Clearpath platforms are waterproof and built with high-quality materials.”
– Marcello Chiaberge, PIC4SeR Centre Team Member
Jackal UGV Gazebo simulation
The core challenges that the PIC4SeR team faced when beginning their project were low-accuracy data from sensors and collaboration among different kinds of algorithms. That is why they required a mobile platform to perform rigorous testing in order to collect vital information on a variety of sensors and develop their brown ROS-based code, which will enable software package deployment. As ROS was central to their project, the PIC4SeR team found it integral that both Jackal and Husky UGV had robust ROS support, allowing them to develop algorithms that could be deployed on different platforms and for different tasks. It provided them with a good level of flexibility and scalability, as well as code reusability and modularity for different applications, such as precision agriculture.
Assembling the Winning Formula
For sensor data collection, the team equipped their robots with an MPU-9250 IMU, a Piksi Multi by Swift Navigation (GNSS receiver), an Intel Realsense d435i, and a 3-axis magnetometer GY-271 HMC5883L. These specific sensors were to keep the overall stack cost under 1000 euros, which would prove the feasibility of a cheaper than industry standard autonomous navigation solution while maintaining reliability and high accuracy. The team’s specific setup allows the autonomous navigation algorithms to have good accuracy, despite the fact that they are cheap compared to other very precise sensors. All of the team’s effort has been spent developing robust algorithms to overcome noisy data in order to use cheap off-the-shelf products and obtain reliable and accurate autonomous navigation in precision agriculture scenarios.
Real-World Testing Data Collection
Alongside the required sensor stack to prove their anticipated low-cost solution, Jackal UGV was particularly integral to their testing. The 4×4 drivetrain allowed them to perform tests on bumpy terrain. As well, the platform was easy to interface with, thanks to ROS compatibility, while having a long usage time (about 4 hours) without charging the battery. As team member Marcello Chiaberge says, “The main motivation is the ROS-support that allows us to easily deploy ROS- compatible code without taking care of low-level communication among different software packages. Moreover, Clearpath platforms are waterproof and built with high-quality materials.“
Comparison of Image Capture and Mask Prediction
The Search for Affordable Robotic Systems Never Ends
Through their rigorous testing, the team has been able to complete an initial autonomous navigation system for vineyards and orchards at a lower price point. In the future, they hope to improve autonomous navigation performances in terms of accuracy and robustness against external disturbances, as well as keep them low-cost.
You can find their work in two publications: “Deep Semantic Segmentation at the Edge for Autonomous Navigation in Vineyard Rows” and “A Deep Learning Driven Algorithmic Pipeline for Autonomous Navigation in Row-Based Crops”. As well, the team received the finalist certificate for the Best Paper Award for Agri-robotics at IROS 2021.
The PIC4SeR team is comprised of Simone Cerrato, Diego Aghi, Vittorio Mazzia, Francesco Salvetti, and Marcello Chiaberge.
To learn more about the PIC4SeR Centre, you can visit their website here.
To learn more about Jackal UGV, visit our website here.
To learn more about Husky UGV, visit our website here.