European Commission logo
English English
CORDIS - EU research results

Article Category

Content archived on 2024-04-18

Article available in the following languages:

An autonomous robot to predict future yields

A technology able to accurately predict future yields, without human intervention, would be a dream come true for vine growers across Europe. If all goes as planned, such technology should be available to them in less than two years.

Food and Natural Resources icon Food and Natural Resources

In a market that’s increasingly driven by quality, VINBOT (Autonomous cloud-computing vineyard robot to optimise yield management and wine quality) can provide European wine producers with a competitive edge. As better forecasts also mean better decision-making, the autonomous robot and its network of sensors can estimate grape yield and relevant canopy features, generate maps, and share such information via the cloud. André Barriguinha, CEO of Agri-Ciência and post project dissemination/commercialisation leader of VINBOT, discusses the project’s outcomes and plans for future work. What are the specific challenges that VINBOT aimed to respond to? VINBOT can run autonomously in the vineyard with no human intervention. It is equipped with a sensor system to allow navigation, localisation and data acquisition. It uses a single camera that collects shots from the canopy. Once it’s done, it uses algorithms to identify the grapes and bunches and estimate future yields. This is something that’s currently not available on the market: There is no existing yield estimation device, so vine growers must resort to manual processes which are very time consuming and not very accurate. Thanks to VINBOT, vine growers will have a new tool to make these estimates as early as possible. How come there was no such solution before VINBOT? Because it’s very difficult to have both a fully-autonomous robot that can navigate inside the vineyard and estimate the yield. VINBOT is bringing just that: autonomous navigation thanks to an on-board GPS receiver and 2D rangefinder; an HMI to define a set of waypoints and the characteristics of the acquisition mission; vine measurement components; a cloud-based software that processes robot’s sensors data to extract relevant information and produce yield maps; and a web application for the end user to access these maps. One of the major problems we faced was trying to identify the bunches in the vineyard, mainly when they are not visible to the camera (hidden by vegetation and/or other clusters). This is a difficulty that we are still trying to overcome by the use of models based on the 3D canopy reconstruction obtained by the Range Finder. Our results showed that canopy features and yield can be estimated by VINBOT platform with an acceptable accuracy. However, the underestimation of actual yield, caused mainly by cluster occlusion, deserves further research to improve the algorithm’s accuracy. We are confident that we can improve the accuracy by conducting further research, either on our computer vision algorithms and on the models to estimate the hidden clusters. We are aiming to put together a second project for that. What would you tell vine growers if you had to convince them of the benefits of using VINBOT? Manual error margins are huge, around 30%. So if they get a technology that helps them reduce this to 10%, it’s a huge plus. VINBOT can estimate their yield; yield maps autonomously and almost in real time; plan cluster thinning needs to prevent excessive production and consequent poor wine quality; as well as improve decision-making regarding planning and organisation. Finally, it can help plan for purchases and grape sales, decide on prices and management of wine stocks, programming investments and develop marketing strategies. Do you already have an idea of what it would cost growers to acquire the VINBOT technology? We don’t think it makes sense for most vine growers to buy a VINBOT, because if it’s only for yield estimation, then they would buy a device that will be in their garage practically all year. We rather intend to make VINBOT available through service providers, and probably wine producers managing large fields. Also, VINBOT is not just the robot: it requires a server for post-image processing, so it would be easier and less expensive for vine growers to use a service provider. Regarding commercial price, the final version of VINBOT would cost around EUR 30 000 including all components —However, as we fine-tune the technology, this price can drop. Speaking of which, VINBOT is cloud-enabled. Why was it important? Because of the large quantity of data to be processed, which makes it easier and cheaper to use the cloud. The algorithm to process the images is stored on a cloud-based processor, so vine growers just need a login and a password to access their results. You also expect vine growers to be able to sell their wine at a higher market price thanks to VINBOT. How? It’s not a direct benefit. If I use VINBOT, I can make better management decisions and indirectly increase wine quality: By optimising yield management and harvest logistics, quality and homogeneity of the fruit, canopy management, cluster thinning and differential harvesting I can plan the production, marketing and wine distribution more efficiently. In theory, this enables wine growers to aim for a higher market price, but it might not always be realistic because of how highly competitive this market is. However, VINBOT can be relevant in decreasing overall production costs, thereby increasing profit margin. What can you tell us about the main results from field tests? We are very happy with the overall robot platform behaviour. We have had a couple of issues with the traction of the wheels on tilled soil and the fact that the current system moves pretty much like a tank, but we are already thinking in implementing a new set of wheels that can turn independently from each other to solve that issue. The next challenge is mainly software and algorithms. We need a more compressive field validation not only focusing on data acquisition of images, but also to help us refine computer vision algorithms and modelling in data processing. This way we will know exactly what we need to do to achieve a under 10-15% error margin regarding yield estimations. Supposing that you get further funding, when do you expect the technology to be commercialised? VINBOT is currently at level 7 of TRL (Technology Readiness Level). We are trying to get H2020 funding, and if we do get it and do the upgrade/validation process goes as planned, it would be possible to see VINBOT on the market in two to four years. We also intend to put our technology to use beyond vineyards, for instance in raspberry greenhouses in Portugal which are interested in the image analysis-based yield estimation. We also have some contacts in the US and are planning to test VINBOT there. Finally, we intend to add more sensors, including environmental ones. VINBOT Funded under FP7-SME Cordis project page Project website



Related articles