Robotics, autonomous machines and image analysis in farming: what is in the field today and what is still in the lab

Farm robots are no longer seen only on trade fair stands; some already work in real fields. Yet the technologies grouped under the word "robot" differ widely in maturity: some have been common for years, some are taking their first commercial steps and some are still in trial plots. Drawing on the European Parliamentary Research Service (EPRS/STOA), the European Commission's Joint Research Centre (JRC), FAO and peer-reviewed studies, this guide summarises what can be bought today and what is worth watching.

Why now?

Two drivers stand out: labour and input costs. The 2023 EPRS/STOA study notes that labour often accounts for 40-60 percent of total fruit production costs in the United States and that the situation in Europe is quite similar. According to the same study, the Spanish citrus sector is under severe pressure because input prices have risen while selling prices for oranges and mandarins stayed flat for about 20 years (EPRS/STOA, 2023). For a labour-intensive crop such as the Kozan orange, this picture is familiar.

FAO reports that autonomous crop robots entered commercial use only very recently, mostly in high-income countries, for example for weeding organic vegetables and sugar beet in France, and that more than 40 start-ups are developing such machines. Adoption by small-scale producers is negligible (FAO, 2022).

Maturity map

Smart spraying: herbicide only on the weed

The idea is simple: herbicide goes only where a weed is, not onto the soil or the crop. Deep learning has markedly improved how well cameras tell weeds from crop plants; the "see and spray" technology, which finds and sprays weeds in lettuce using machine vision, is a well-known example. An autonomous orchard sprayer developed in Brazil can reach 15 km/h and stops spraying where there is no vegetation (EPRS/STOA, 2023).

How much can be saved? According to a peer-reviewed review, patch spraying saved at least 50 percent of herbicide in various crops without additional weed control costs in the following years. The same review points to a major obstacle: most systems use their own IT protocols, so sensors and implements from different manufacturers cannot be connected; plug and play standards would speed up adoption and lower operating costs (Gerhards et al., 2022). The rules for spraying with drones are covered in a separate guide on our platform.

Image-based disease detection: the gap between the lab and the orchard

A deep learning model trained on 54,306 leaf images taken under controlled conditions reached 99.35 percent accuracy on a held-out test set in distinguishing 14 crops and 26 diseases. The same model dropped to around 31 percent on images taken under different conditions. The authors note that the model is limited to single leaves on a uniform background, while many diseases show symptoms outside the upper leaf surface (Mohanty et al., 2016). EPRS/STOA also stresses that training deep learning algorithms requires large numbers of labelled images from different locations, seasons and varieties, and that spectral imaging is promising for pathogen detection, especially in greenhouses and vertical farms (EPRS/STOA, 2023). In practice, these tools are useful as early warnings, but a diagnosis still needs confirmation by an agronomist or a laboratory.

Why are harvesting robots still difficult?

Selective harvesting means finding the fruit, approaching it and detaching it without damaging the plant. A review of 50 harvesting robots found average success rates of 85 percent for localisation, 75 percent for detachment and 66 percent for the whole harvest; the average cycle time was 33 seconds per fruit and none of the robots had been commercialised (Bac et al., 2014). The economics are also cautious: a review of 18 studies shows that all found some scenarios in which automation was profitable, but most relied on prototype data or on the assumption that design targets would be met, and all were carried out in developed countries (Lowenberg-DeBoer et al., 2020).

Costs, regulation and other limits

• Perceived price: In a JRC survey of 1,444 farmers in nine EU countries, including Greece, 77 percent of respondents considered fully automatic field robotics (hoeing units, weeders) expensive or very expensive. Self-financing covered less than 40 percent of adoptions of these robots, while leasing or renting accounted for 27 percent (JRC, 2025).
• Supervision rules: In some countries autonomous machines must be supervised on site at all times, in which case conventional equipment may be the better option. Speed restrictions can also undermine profitability (FAO, 2022).
• EU machinery law: Machinery Regulation (EU) 2023/1230, applicable from 14 January 2027, defines a remote supervisory function for autonomous mobile machinery: it only allows the machine to be stopped, started or moved to a safe position, and if the function is not active the machine cannot operate (EU, 2023).
• Skills: In 2016, 57.9 percent of EU farm managers were aged 55 or over and 68.3 percent had only practical experience; EPRS/STOA therefore calls for a strong extension network and training (EPRS/STOA, 2023).

Starting points for farms in Kozan and Larissa

1. Start with mature technology: with GNSS guidance and section control, input savings from reduced overlap are almost immediate (FAO, 2022).
2. Share expensive equipment: EPRS/STOA notes that high costs can be shared through farmer associations, cooperatives or service providers.
3. Ask for a demonstration in your own orchard or field before buying; weed density, row spacing and soil condition determine the result.
4. Ask about data and compatibility from the start: which data format the machine uses, whether it works with another brand's sensor and who owns the data should be in writing.

Sources

1. FAO, 2022. The State of Food and Agriculture 2022: Leveraging automation in agriculture for transforming agrifood systems, Bölüm 2: The state of digital automation technologies and robotics in agriculture (accessed 29 September 2026)
2. Avrupa Parlamentosu EPRS/STOA, 2023. Artificial intelligence in the agri-food sector: Applications, risks and impacts (PE 734.711) (accessed 29 September 2026)
3. Avrupa Komisyonu JRC, 2025. Tur Cardona vd., The state of digitalisation in EU agriculture: Insights from farm surveys (JRC141259) (accessed 29 September 2026)
4. Bac, van Henten, Hemming ve Edan, 2014. Harvesting robots for high-value crops: state-of-the-art review and challenges ahead, Journal of Field Robotics 31(6) (accessed 29 September 2026)
5. Gerhards vd., 2022. Advances in site-specific weed management in agriculture: A review, Weed Research 62(2) (accessed 29 September 2026)
6. Mohanty, Hughes ve Salathé, 2016. Using deep learning for image-based plant disease detection, Frontiers in Plant Science 7 (accessed 29 September 2026)
7. Lowenberg-DeBoer, Huang, Grigoriadis ve Blackmore, 2020. Economics of robots and automation in field crop production, Precision Agriculture 21 (accessed 29 September 2026)
8. Avrupa Birliği, 2023. (AB) 2023/1230 sayılı Makine Tüzüğü (EUR-Lex) (accessed 29 September 2026)