This year, numerous reports have posited that agriculture stands on the cusp of another technological revolution, with drones and robots - now frequently used in research and field trials - set to revolutionise field work in the near future, allowing for ‘ultra-precision’ agriculture.

Robot weeding implements are already available, and the next generation of robot weeders is expected to use huge troves of data and ‘deep learning’ techniques to improve their work.

But what’s the next step? This great leap forward in technology will change robots from machines following a set of directions, or processing from a set of pre-established options, to ‘learning’ new solutions; having been given a set of basic rules by a programmer, the machine will learn to make the best choices for the job it does.

In agriculture, the current focus for machine learning is on monitoring and forecasting - jobs that can be done every day to absolutely optimise crop growth. There is scope for roaming robots to tackle pests, diseases and weeds, and with much more precision and less environmental impact than current methods allow. The process at work will involve sensors feeding necessary information about crops or herds in real time to the machine, which will analyse and continually cross-check data with added information already inputted (like the cost and availability of inputs, weather forecasts, disease stats and so on) and then take the best possible action based on its knowledge.  

These machines will require a huge amount of data. Learning machines on farms will require information from multiple sensors in the field. These sensors already exist, though mostly they have been designed with human interpretation in mind. They include collars that feed back information on cattle behaviour and temperature (from which a computer algorithm can identify areas of concern), to probes that can look at weather in certain specific areas and cameras that can identify signs of disease in plants.

Factors limiting the widespread deployment of robots in farm fields include power (charging and battery life) and the strain of outdoors work in all weathers - the need to cope with wind, rain mud and dust. However, there are already groups working to overcome these obstacles, and promising solutions include the NH Drive, developed by the CNH Industrial group and the Bonirob, developed by Bosh.

Artificial intelligence software, that will take this next step, allowing machines to process data, make decisions, and learn, is under development, but there is still some way to go. Some software, like that developed by the UCM (Universidad Complutense of Madrid), can take decisions but can’t process data in real time, like a human can, and can’t autonomously put its decisions into practice, though these issues should be ironed out in the next few years.

By combining hardier machines, better data collection techniques and highly developed software, the next level of computer learning doesn’t appear to be too far away, and it’s easy to imagine how useful it will be to tomorrow’s farmers, who are facing pressures from climate change, to plateauing yields to resistance to chemistry in weeds and pests.
 
This is adapted from an article by Pascal Cochelin, a digital business leader, which appeared on Terre-Net. Pascal believes that, despite all the high-techery poised to revolutionise food production, farmers will still be essential in the farming process; in overseeing machines, planning, ensuring seeds take and harvests are brought in. He says, “Though machines with artificial intelligence are surprising in their adaptability and prospects for improvement, they still lack a very human factor: common sense.”