Robots and automation for cropping


Robotics and automation are going to happen within a generation. Once the concept is accepted adoption rates will be high and the costs of the technology will come down. Academic research often lacks a route to commercialisation so incentives should be offered to encourage this. The UK is lacking in innovative, engineering entrants in agriculture so we need to modernise the way agricultural engineering is taught at university, college, school and even primary levels. Threats about liability are holding back autonomous technology. Autonomous machines will be safer, more consistent, more efficient and offer levels of plant agronomy never thought possible - and the technology is ready to go.

Autonomy in agriculture

The term autonomy for agriculture refers to the full or partial replacement of human interaction with the aim of achieving greater accuracy, consistancy, safety and reliability.

There are thousands of tasks which can be defined as autonomous which have crept into everyday operations without real acknowledgement of their presence; eg automatic gearbox, traction control, automatic braking systems, electronic stability and cruise control. Since the first mechanised tillage, fundamentally tractor units are much the same now as they ever were; a power source (horse, ox, tractor) and an implement, usually mercilessly dragged through the ground.

The current trend is to increase the size of the implement to be more efficient and cover greater swaths of land in single passes, requiring more and more horsepower, causing higher levels of compaction, which then requires more energy to rectify the damage. This is a vicious circle resulting in larger machines using more fuel, metal, machine hours and man hours to repair its own damage.

Robotics in agriculture is not a new concept, it has been a target for agricultural visionaries for many years. The first report I came across was from the February 1934 Modern Mechanix Journal. It showcased the concept that a farmer could be separated from a machine’s field operation. Between the birth of the concept, and 2005 there was little progress in the automation of arable robotics or indeed any automotive sector. Most manufacturers have tried some concepts involving wires, radios and other ingenious systems but all had the same problem: intelligence, or lack of.

In 2005 the Defense Advanced Research Projects Agency (DARPA) grand challenges led to the advent of many autonomous vehicles in a range of industries including aviation, mining and agriculture. The most famous provided the seeds for the Google self-driving car which is currently driving around San Francisco.

The industry seemed to have get stuck at the point of auto steer guidance and improving the functionality and usability of this - not pushing towards the next milestone of true automation and the partial or complete removal of the human presence in a tractor cab.

The dairy industry across Europe has accepted robotics over a short period. It is commonplace, and still a growing market, with the majority of milk produced in Holland, Denmark and Sweden now having passed through a robotic milker. What fascinated me was that even though the machines operate in a semi controlled environment, the variables of an autonomous object dealing with a living, moving animal would pose a technical nightmare for any system. I was intrigued by how, firstly, it was technically possible and how that technology may be applied to broadacre arable; and secondly how a herdsman trusted such a machine with his livelihood; and finally how legislation, red tape or liability didn’t cripple adoption of such systems.

Drivers to Automation

Whilst in Australia I witnessed weed spraying across vast hectares of land. Yet approximately only 5% of the area needed chemical. However the operation required man time, machine time, and vast amounts of water and chemical. A small lighter vehicle, performing the same task but with intelligence, could offer huge savings. With multiple machines plus 24/hour operation - suddenly more can be achieved in a shorter time.

Commercial products are already available for the autonomous removal of bulk grain from fields through modified tractors and chaser bins. Once again this offers labour saving and safer operation. Removing the variability of human performance can increase efficiency and consequently fuel usage.

All the autonomous operations I have seen make safety a key part of the design. Google’s autonomous car has now travelled 300,000 miles in full autonomous operation accident-free, although it has had two accidents when driven by a human. Autonomous mines in Australia have seen a reduction in injuries because operators don’t have to climb in and out of large vehicles.

Currently there are no guidelines, tests or accreditations which autonomous agricultural equipment or vehicles must abide by.

Being able to monitor so many aspects of individual specific plants throughout their growth stages is a very powerful tool. Possibilities include: selective breeding programmes, picking plants with specific favourable traits such as drought resistance, thicker stem for reducing lodging risks, and even monitoring leaf area to perhaps develop breeds ideal for lower light levels. Teamed with laser weeders a very rapid breeding programme can be generated. The shift from field level operations to individual plant level operations will result in plant optimums and not field optimums raising overall yield whilst reducing inputs.

Social and economic impact
I looked at industries which have already adopted such systems, for example manufacturing, transportation and aviation. I discovered an array of situations where it has had a negative effect. Automation can induce new forms of stress due to information overload, skill-degradation, boredom, complacency and over-reliance on the system. Most failures are caused by poor training, poor human computer interaction or lack of understanding of responsibilities. In due course it is inevitable that the introduction of robots will make some jobs and skill sets redundant; however it will create other opportunities.

A skilled workforce of technically literate staff will be required for the lowest levels of operation, whereas further up the chain highly educated engineers will be required to design the systems and control the approach which will be implemented. For situations where a large workforce will be required to operate side by side with autonomy - such as fresh produce - management styles will have to change from a dictative stance to a bi-directional understanding to ensure both humans and machines are operating efficiently.

Trials have shown that low-skilled workers react negatively toward the implementation of robots, perceiving them largely as threats to their job security. High-skilled workers reacted more positively and perceived the implementation as providing opportunities to expand their skills set and knowledge.

Although automation will result in less human input, it will cause an increase in technical support roles. These roles will require knowledge in subjects not currently within any agricultural college’s syllabus and currently the skills set would only be available from a few companies across the UK.

If the new systems are to be readily adopted the responsibilities will in the short term fall upon the manufacturers to effectively train and educate the service field staff. However, having general engineering knowledge is a key part of understanding the processes and systems in place.

Global Navigation Satellite Systems

On the other hand at higher design levels, software programmers, engineers and project management will need to complement the technical skills they already have with fundamental agricultural practices. I believe these will be the hardest roles to fill.

The inconvenient truth

Machines will inevitably be designed and built by a range of manufacturers with a range of different ideas and concepts; this will cause a technology and standards battle. Extensive work is needed defining standards and recognising the requirement for communication between machines.

Furthermore there are major questions around the liabilities for autonomous systems. The only place to seek any hint on direction for these questions is the automotive autonomy industry and even they are struggling to build a structure with transparency. Currently any system which is legally on the road must have a human who at any stage can take control of the system. However this is already being challenged due to the vagueness of the definition of a driver.

The vision

All that seeding requires is for a seed to be placed in the ground - perhaps a small amount of micro tillage at placement to break the surface, but nothing more. Low ground pressure implements working long hours unsupervised, using its higher resolution of operation to utilise background maps obtained from previous sensor information for variable seed rates and fertiliser placement, can do the job.

With slower operational speeds seeds can be intelligently placed in uniform patterns, not only suppressing weed growth, but also achieving an even, light distribution, even fertiliser extraction and space for the plant to grow. Herbicide applications can be reduced between 50% and 100%.

There is always going to be a bulk element required for both seeding and the harvest. This point has often been overlooked by researchers. A solution is to have bulk traffic pathways with the same principles as controlled traffic. Inter machine co-operation, loading and loading of bulk product, can be done from the bulk traffic pathways. Scouting and nurture robots having minimal weight can operate in much smaller, narrower paths allowing access to the whole cropped area.

Scouting & Nurture
Due to the removal of time constraints, the advances in sensor technology and the reliability of GNSS steering, machines will be able to acquire very high resolution information about the current crop health and nutritional situation. This in turn can be used to make management, or even autonomous, decisions as to what the appropriate action will be.

The large amounts of data that this will generate will require an operator, manager, agronomist or even a new job title to interpret the information. Systems such as BoniRob have proven the concept of creating plant level databases containing an array of variables to base decisions upon. This in turn needs to be supported by a mechanism of providing a physical application of the plant level agronomy. High resolution spray application and inter/intra row weeding will reduce the need for blanket chemical treatments.

Some of these tasks can be completed on the scouting pass, others with require a set mission to complete. Complemented with arising technologies such as laser and inter row mechanical weeding, the overall application and treatments will be reduced.

Once a mission has been created the vehicle must be prepared with the relevant tool or chemicals. This should also consist of an operational safety check ensuring correct operation of sensors and equipment. Once the vehicle is prepared it can be released to perform its operation autonomously. In the short term the goals will be specific to an individual machine; however, with better interoperability and inter vehicle communicational standards, a general plan may be released and the machines may design and execute the mission themselves.

Utilising a mixture of real-time nutrient analysis and background layer maps from earlier scouts, treatments may begin. Multiple machines may work on a single task.

Perhaps the most damaging time for soil structure is harvest. I do not believe smaller machines will be suited for harvest operations in arable situations. However in vegetables or row crops smaller machines will be able to pick individual produce based on the latter’s current attributes, and only harvest that which is ready.

Why has this not already happened?

The scenarios laid out above are all technically possible today, with no new technology but simply by applying several individual technologies together. The missing piece in the majority of projects witnessed is a route to commercialisation. European academic developments tend to stop once a particular funding stream has finished. This is not so true in Japan and Australia, where the workshops had an entire history of developments, with robots over 20 years old in Japan. There was a culture of commercial support provided by machinery, implement and electronic manufacturers working in conjunction to solve problems and offer a route to market.

Nearly all the global manufacturers I visited viewed the demand for robotics in an unenthusiastic way - there is an apparent lack of interest in changing away from high horsepower machines.

From James Szabo’s, Nuffield Farming Scholarship report “Autonomy in Agriculture.”