Periodic Reporting for period 3 - ROMI (RObotics for MIcrofarms)
Période du rapport: 2020-10-01 au 2022-07-31
ROMI developed an open and lightweight robotics platform for these microfarms, by proposing for the first time a weeding robot, a farmer’s dashboard with a cablebot, and a 3D plant imager. With these tools, we are able to assist these farms in weed reduction and crop monitoring. It reduces manual labour and increases productivity, notably through advanced planning tools. For instance, thanks to ROMI’s weeding robot, farmers could save 25% of their time. This land robot and the cablebot also acquire detailed visual information about the crop beds that is analysed and visualised in our farmer’s dashboard application. They produce an integrated, multi-scale picture of the crop development that helps the farmer monitor the crops to increase efficient planting and harvesting. For this, ROMI adapted and extended the state-of-the-art for land-based and airborne monitoring tools to handle small fields with complex layouts and mixed crops.
To achieve this, we: (i) developed, and plan to bring to the market, an affordable, multi-purpose, land-based robot, (ii) developed a weeding app for this robot that is adapted for organic microfarms, (iii) applied advanced 3D plant analysis and modelling techniques to data acquisition, (iv) integrated AI-based analysis techniques for detailed plant monitoring, (iv) integrated these techniques also in an cablebot for multi-scale crop monitoring, (v) proposed an affordable 3D scanner for plant phenotyping and (vii) tested the effectiveness of our solution in real-world field conditions at the Chatelain market farm.
The Plant Imager is a physical phenotyping station where plants of interest are placed to be automatically imaged with a moveable RGB camera. We also provide a suite of integrated analysis pipelines performing automated 3D reconstructions. The Plant Imager and its software modules provide a comprehensive environment to perform automated, quantitative, precise, 3D phenotyping of a single plant shoot architecture. It is being finalised in controlled laboratory settings for a future dissemination towards research laboratories. However, its use in controlled farming environments, such as greenhouses, is envisioned as a mid-term application.
The Cablebot is a camera that moves automatically along a cable to collect images of crops. The device is associated with the Farmer’s Dashboard, the web interface to display and analyse the image data. The software module maps crop rows, locates and identifies individual plants, and extracts their growth curves. As part of the ongoing research activities, we also investigate crop growth prediction or the optimization of crop placement.
The Romi Rover assists vegetable farmers in maintaining vegetable beds free of weeds. It regularly hoes the surface of the soil and thus prevents small weeds from taking root. It can do this task mostly autonomously and requires only minor changes to the organisation of the farm. It is designed for vegetable beds 70-110 cm wide and for crops up to 50 cm high. A weekly passage of the robot is required to keep the weed population under control.
The research activities within the ROMI project are also highly productive. We are evaluating novel strategies for more open-ended learning in robotics. This includes learning through curiosity-driven and reinforcement learning so that the robots can intervene robustly and efficiently when dealing with complex and unpredictable situations on the farms.
The research on virtual plant models has the potential to significantly speed-up and improve the training of AI systems. These AI systems are needed in the automatic analysis of plant images to detect their organs (such as the flowers, fruits, and stems) and are an essential component in any advanced robotic device.
ROMI devices also support the development of plant research, which implies analysing plants’ characteristics (phenotyping). Automatic phenotyping devices currently commercialised tend to be of a high throughput and expensive, thus reserved to phenotyping professionals. Yet, the 3D plant imager is medium throughput and more affordable. It can therefore be used by other types of scientists, including those studying small/medium size plants. The Farmer dashboard could also foster plant research, notably in experimentation platforms. Scientists doing varietal selection, for example, closely monitor plants’ growth by hand. However, devices automatising plants’ data collection are more rapid and reliable than manual data collection. The 3D plant imager and the Farmer Dashboard can therefore support plant research in laboratories and experimentation centres.
Finally, the ROMI tools (hardware and software) are available under free licence to have the widest impact possible. This model facilitates access to results to small farms worldwide and facilitates the collaboration between farmers, scientists, engineers, and industry. On top of fostering the development of sustainable farm models and plant research, ROMI therefore developed, and will continue to develop, synergies between them.