We help our customers develop breakthrough products and services within their market.

The last two years have shown an incredible, and game changing growth in the capabilities of machines, and automated systems. Stories about humans ‘going the way of the horse’, or drastic changes to our financial system being needed abound. Some of this is fantasy, but it’s clear that agriculture is facing both a need to automate, and a push of immature technology from start-ups. For instance, last year a hectare of barley was sown, tended and harvested completely autonomously.

However, farms are complicated places. For instance, fruit and salad harvesting is not an unskilled job which can be done by the cheapest, most transient labour. It can take several seasons to learn to pick with both speed and high product quality. Similarly, the skill and intuition needed to get a good yield from arable crops requires both local knowledge and an ability to evaluate the hundreds of things which can affect a particular season. Clearly, automating these tasks can’t happen overnight - the machines still have to gain the skills and knowledge required.

Looking into this commercial maelstrom – a desperate need from agri-business combined with a well-funded push from technology firms – which robots will win? In the short term, it may be the first to market or the best funded, but over time things will return to fundamentals. And this means the winners will be the ones which began with the business case and build a solution which fits.

For example: Harry Ferguson popularised the three point hitch – this had the immense advantage of transferring the resistance from the plough into downward force on the drive wheels. This not only prevented the tractor from flipping backwards if the plough met a stone, but meant the tractor could be lighter, cheaper – and cause less soil compaction. This became standard – Ferguson even convinced Henry Ford to adopt it.

Compare this to drones in agriculture. Although they can, in theory, survey large areas at low cost, early adopters have found the need to wait for a pilot, the right weather conditions and the required post processing is hard to justify. This may be bearable for a detailed field trial, but it’s a long way from giving any advantage to broadacre crops.

Here are three key observations:

  • Robots succeed where robots are welcome. This isn’t stated to appease the ‘they took our jobs’ lobby – it’s way to state that cooperation with humans will be required. If the machine is easy to run, unload and look after then employees will enjoy working with it, and become better at driving it. If it’s dangerous or has a cryptic user interface, staff will avoid it or it may ‘get broken’.
  • Developers and users need to agree KPI’s. This isn’t to appease ‘bean counters’, but a powerful way to ensure engineers are optimising for the right thing. For instance, a machine which picks 98% of your fruit sounds exciting – until you realise it’s indiscriminately picking green ones and destroying next week’s crop. These KPI’s can take the form of a cost (or penalty) function which includes such things as the value of the crop, the opportunity cost of breakdowns and the labour required to run it. Thinking in this way will also bring the business case to the front: for instance a ‘good enough’ machine which is cheap to own may outweigh something much more polished.
  • Allow sufficient time for training. The breakthroughs in machine learning have come from improved training techniques – these require not only huge amounts of CPU time (which is easily bought) but beautiful training data. Often, this training data has been labelled and curated by humans using images the robot itself has collected. It will take time, effort and people with a real stake in success to collect and use properly.

It’s childishly animist to say that ‘humans have to teach machines’. But, much of robot and machine learning development is about following a path to an ‘optimum’ outcome. The key job of agronomists, farmers and engineers is to agree what that optimum should be – that way the technology will drive on.

Simon Jordan
Senior Sensor Physicist

Working in our sensing systems group, Simon specialises in navigation and communication. Before joining Cambridge Consultants, he spent ten years at Teledyne TSS, working on projects including electromagnetic pipe tracking/survey systems, ship steering systems, marine motion sensors, and the development of high grade inertial navigation systems.