Harper Adams delivers first autonomous farm harvest

Researchers from Harper Adams University in the UK have partnered with startup Precision Decisions, in a successful trial to grow barley using autonomous farm machinery. This should be a shot in the arm for agricultural automation, which is looking for ways to improve its margins – and cater for rising global populations.

The robotic tractors and harvesters were augmented by surveillance drones, in a project called the Hands Free Hectare (HFH) project – the first to achieve a barley without a single person stepping foot in the field itself. The crop is going to be used to make beer, and while still in early stages, the approach has great potential.

Although the project was successful, after completing the harvest in September, the yields were much lower than what a regular farmer might achieve using conventional equipment. The HFH managed to generate 4.5 metric tons of barley per hectare – below the 6.8 tons you might expect to generate on a regular farm, a 33% difference.

All the machinery used in the HFH is readily available to farmers. The Iseki tractor and Sampo harvester are common models, while the navigation system is open source and normally used by drone pilots.

Getting the project set up cost around £200,000, supplied using funding from the UK government, making it probably the most expensive hectare of barley crop ever produced. Given that most of these were mostly capital costs, purchasing automated tractors and harvesters, next year’s harvest shouldn’t be quite so expensive – although this is a research project, rather than a commercial venture.

The project’s focus was to prove the concept possible, rather than do better than a regular farmer. Riot spoke with HFH project leader Kit Franklin, who said the low yield was a result of time pressures faced by the team, who had to get the machinery ready for April to start drilling the field and get the seeds in before it was too late.

Franklin said that the autonomous driving system for the tractor had not yet been refined, and consequently the coverage of the field was around the 80% mark. Conventionally, a farmer would hit 100% coverage. Franklin said they were aware of the issues with precision but wanted to get the project off the ground for this year and be the first to deliver an automated harvest. That 20% gap suddenly makes the 33% difference quite a lot smaller – so this looks like a very good first try.

Another factor affecting the crop yield was the process of agronomy (deciding which chemicals to use on a crop). Normally a farmer would regularly inspect the field and take multiple samples of the soil to test its condition. Franklin and the team were reliant on samples taken by robots and images taken by drones to access the field, and using these alternative approaches for the first time also impacted on the yield.

Franklin said that both the automated driving and agronomy systems have both been improved since the first harvest. An extra IoT sensor to detect disease level will be added from Bayer Crop Science, a subsidiary of Bayer Chemicals, that should improve the agronomy readings.

The automated driving system has also been improved and Franklin expects that the seed coverage will now be 100%. HFH is already attempting its second harvest and hopes its improved system will now include weather data, which will mean the yields are commercially viable.

There are hopes that automation will enable the use of smaller, lighter vehicles, which could make sowing and harvesting more precise – and importantly, reduce damage caused to land as they move across fields. It is not clear whether farmers would adopt such a model, of having multiple specialist machines, rather than fewer general-purpose vehicles – but this is an evolving field.

The automated driving system used an open source drone platform navigation, that is purely GPS based. Franklin said the navigation software was incredibly cost effective to implement, and now it is better optimized, it should deliver the commercially satisfactory performance. Franklin said that the team are already exploring how they could use LiDAR and Machine Vision technology in their automated tractors further.

Franklin described how the aim of the project was to bring greater attention to the area of automated farming, and to be the first internationally to achieve an automated harvest – and set the bar for the industry. It seems that they have achieved this.

In terms of commercialization of autonomous tractors and farming systems, the race to be the market leader is well underway. Japanese tractor manufacturer Kubota started demonstrating a fleet of autonomous tractors last year. Domestic competitor Iseki is said to be hot on the heels of Kubota, and should be launching autonomous tractors this year.

Japan looks to be the most promising market for autonomous tractors. The average hectare per tractor driver has increased from 18.9 in 2000 to 30.1 today. There aren’t enough tractor drivers to service the market, and with a shrinking population, the demand for autonomous tractors is driving innovation in Japan.

Franklin believes that within the next 5 years, autonomous farming system are going to see major commercial deployments. Specialist farmers growing high value crops like lettuce, which require lots of attention and management but have higher margins, are likely to be the fastest adopters of autonomous systems. Vertical farming is another promising venture, but lies far outside the AI and ML area.

Within 15 years, Franklin sees autonomous tractors being widely adopted, which is likely to reshape the nature of work in the sector, potentially displacing part of the traditional workforce – although that concern doesn’t ring true for places like Japan.

Automated farm processes reduce the typical constraints made of an agricultural workforce. Monitoring and activity can take place 24 hours a day,  and robots could potentially continuously monitor and analyze the condition of crops, in a way that would be impractical and laborious for humans. This should result in a better harvest.

Other significant autonomous farming efforts have come from John Deere, which acquired Silicon Valley AI firm Blue River for $305m. Blue River has several farm tools, from an automatic precision weed-sprayer to a device that trims lettuce at scale, and a software drone to analyze crops.

John Deere already has some autonomous driving technology, like tractors that can steer themselves via help from GPS signals, while image sensors determine the quality of grain during harvesting. John Deere says the acquisition of the Blue River will allow future tractors to understand each individual plant crops like lettuce and cotton, two areas Blue River has already showcased.

The advantage of introducing the image recognition system is that it allows for much more efficient farming – something often called Precision Agriculture. It contrasts the current approach that involves blanketing entire fields with pesticides and fertilizers– which is wasteful. It allows the farmer to use precise amounts of much stronger chemicals, achieving a better result. Blue River said its precision spray system, known as ‘see and spray,’ can reduce the usage of chemical by as much as 95%, and slash the labor cost of weeding.

Researchers from a Bosch startup called Deepfield robotics are pushing a different approach that would move agriculture away from chemicals entirely. Deepfield has built an autonomous robot that can manually weed fields, using a AI machine vision system and an automated weeding tool – removing the need for some pesticides entirely.

Such chemicals can damage the same crops they are supposed to protect, as well as damaging the soil on which the crops are grown over time. These chemicals can also become very expensive for farmers. The Deepfield system has the potential to remove the cost of spraying chemicals entirely – although currently the project is still just being demoed.

Robots are likely to get much more affordable and efficient, and as automakers more intensely pursue autonomous driving and machine learning capabilities, they will be able to leverage these efforts to further develop and reduce the costs of autonomous systems applied in agriculture.