Periodic Reporting for period 2 - Agri-Think Robot (A Disruptively New Approach to Weed Management Shaping the Future of Sustainable Agriculture)
Berichtszeitraum: 2024-01-01 bis 2024-12-31
Weed control is a big problem in both organic and regular farming, and it has a big impact on the quality and amount of crops. Right now, people use herbicides and manual methods to deal with weeds, but these methods have problems related to the environment and labor. Herbicides can be harmful to health and the environment, and weeds become resistant over time. Manual removal is hard work, expensive, and faces challenges related to labor availability and rules. With more people to feed and a growing demand for food, there's a need for new ways to manage weeds that are efficient and sustainable.
Odd.Bot has come up with a solution called the Agri-Think Robot. It's a lightweight robot that works semi-autonomously and is designed to manage weeds in rows of crops. This technology is especially useful for crops like carrots, onions, and chicory, where plants are close together. The robot uses a camera to look at the crops and weeds and figure out which is which. Odd.Bot uses groups of robots in each field to quickly and efficiently remove weeds, making the best use of the operator's time.
Machine learning is a crucial part of how the Agri-Think Robot works. The robot learns to tell the difference between crops and weeds by using a database with thousands of pictures. Farmers start the database by labeling pictures manually, and then machine learning takes over and keeps improving the database over time. This database is available online, allowing for real-time adjustments and retraining of the robot.
Odd.Bot's goal is to speed up the shift to sustainable farming by offering an alternative to using chemicals or manual weeding. One of the key features is its quick learning ability through live labeling, allowing instant sharing of knowledge among the group of robots. This continuous learning process makes the robot better at recognizing weeds and crops over time, and the knowledge can be shared with other groups of robots.
The project has seven parts, or work packages:
Project Management
Commercialization and Communication
Field Testing & Preparing for Trailblazer Demonstration Program
Make Weeding Robot Ready for Series "0" Production
Software & Operations
Big Data for Precision Farming
Preparation and Scale-Up for Production and Market Introduction
These packages cover everything from managing the project and introducing the robot to the market to making technical improvements, enhancing software, and getting ready for full-scale production and market launch. The project takes a comprehensive approach, making sure every aspect is considered for a systematic and effective move towards sustainable weed management in agriculture.
Odd.Bot has come up with a solution called the Agri-Think Robot. It's a lightweight robot that works semi-autonomously and is designed to manage weeds in rows of crops. This technology is especially useful for crops like carrots, onions, and chicory, where plants are close together. The robot uses a camera to look at the crops and weeds and figure out which is which. Odd.Bot uses groups of robots in each field to quickly and efficiently remove weeds, making the best use of the operator's time.
Machine learning is a crucial part of how the Agri-Think Robot works. The robot learns to tell the difference between crops and weeds by using a database with thousands of pictures. Farmers start the database by labeling pictures manually, and then machine learning takes over and keeps improving the database over time. This database is available online, allowing for real-time adjustments and retraining of the robot.
Odd.Bot's goal is to speed up the shift to sustainable farming by offering an alternative to using chemicals or manual weeding. One of the key features is its quick learning ability through live labeling, allowing instant sharing of knowledge among the group of robots. This continuous learning process makes the robot better at recognizing weeds and crops over time, and the knowledge can be shared with other groups of robots.
The project has seven parts, or work packages:
Project Management
Commercialization and Communication
Field Testing & Preparing for Trailblazer Demonstration Program
Make Weeding Robot Ready for Series "0" Production
Software & Operations
Big Data for Precision Farming
Preparation and Scale-Up for Production and Market Introduction
These packages cover everything from managing the project and introducing the robot to the market to making technical improvements, enhancing software, and getting ready for full-scale production and market launch. The project takes a comprehensive approach, making sure every aspect is considered for a systematic and effective move towards sustainable weed management in agriculture.
Odd.Bot has focussed in 2023 on finalising the Odd.Bot robot from a mechanical and mechatronic perspective: The mobility module, the 2 delta arms, the energy supply module, the detection module and the design have been improved/ adapted and tested in season 2023. Odd.Bot focussed on carrots in 2023. Odd.Bot also collected data from Onion fields and weeding in Onions as the first test. Out of the test results, it became clear that not all KPI’s were met (yet). KPIs are defined as weed removal accuracy, capacity, and crop damage. The Weed removal capacity was a milestone realised as planned before the end of May 2023 (see Milestone 2). The weed removal accuracy was measured at 85% or above for growth phase BBCH 11 and 12 for carrots which is on target. The weed removal accuracy for growth phase BBCH 13 and above and below BBCH 10 for carrots stayed behind, resulting in mitigation to improve the database and start a new 2024 mechanical plant fixation project to “look under the crops”. Crop damage was in 2023 only initially estimated and will be measured in 2024.
Odd.Bot also started an improvement of the AI database: Annotated database (initially 2517 images of weeds and crop) extended with 6322 new weed images and 8255 crop-only images, resulting in better weed removal accuracy.
Odd.Bot further made the robot autonomous driving to eliminate the permanent operator, making it also (from both farmer's and Odd.Bot's perspectives) a commercially more attractive proposition. Part of autonomous operation is also self turning on the headland which was realised in 2023. This can be seen in video https://youtu.be/asolWmguAh4(öffnet in neuem Fenster)
Odd.Bot also started an improvement of the AI database: Annotated database (initially 2517 images of weeds and crop) extended with 6322 new weed images and 8255 crop-only images, resulting in better weed removal accuracy.
Odd.Bot further made the robot autonomous driving to eliminate the permanent operator, making it also (from both farmer's and Odd.Bot's perspectives) a commercially more attractive proposition. Part of autonomous operation is also self turning on the headland which was realised in 2023. This can be seen in video https://youtu.be/asolWmguAh4(öffnet in neuem Fenster)
New for Odd.Bot is autonomous driving on a flat field (without ridges as guidance, just using crop location and GPS to drive the robot) and driving as stand alone without permanent operator. This is impacting the total addressable market for Odd.Bot. This is not a result beyond the state of the art as the used technology in general is already known.