I’m attending the Agritech East pollinator on Tuesday, all about the potential of precision robotics in agriculture. My talk is about automated harvesting, and where robotics is and isn’t relevant. It’s clear that agriculture has many tasks, currently done by humans, which are amenable to some level of automation. That human labour is becoming more expensive and scarcer, especially as more people even in developing countries are demanding more highly skilled jobs. Robots are becoming more capable too, and are appearing in all sorts of new places. So why hasn’t it happened everywhere already? First, let’s be clear what we mean by automation in this context. Farms have had some level of automation for many years – but it’s been focussed on broad-acre crops, things like potatoes and corn where you don’t need accuracy or intelligence to harvest. I’m not talking about removing the human completely; there’ll always be some jobs that are just to hard to automate, and I’m not talking about replacing a human – you can banish thoughts of anthropomorphic androids walking around fields picking apples and sowing seeds. Instead, I’m looking at the middle ground – tasks that can’t be done by a combine harvester, but still simple enough that automation can be cost effective.
And it’s that “cost effective” that’s a key point. For harvesting to be worth automating, it’s got to be commercially feasible as well as technically feasible. The scale of the opportunity is large enough – for example, of the approximately 20 billion apples grown in the US each year ALL of them are picked by hand. But automation is expensive, both in terms of unit cost and development cost. It might cost tens of thousands of pounds to build a single piece of automated harvesting equipment, but it’ll be millions of pounds to develop it. Even looking at a five year time scale, you’ve got to either replace a lot of labour or have many installations before it starts to make sense.
One of the things I find most interesting about the idea of automating apple picking, for example, is that it’s only the first 18″ of the apple’s journey that doesn’t have any automation. Once it’s out of the tree and into the box, it can be carried by machines, sorted and graded in factories, transported to warehouses and delivered to supermarkets or the customer’s door – all of those processes have some level of automation. But the first step from tree to crate has – up until recently – been entirely manual. That’s the space where we’re just starting to see automation starting to be technically feasible, and make sense commercially.
We’ve done some work on machine vision, as that’s one of the keys to solving automation in messy environments like agriculture, that’ll be talking about too. This includes deep convolutional neural networks, which while very useful on a whole host of machine vision problems, weren’t ideally suited for our fruit picking challenge.
So more automation in agriculture is, in general, worth investing in, BUT:
- The business case must make sense which typically means many installations are required
- You have to start with the problem and work out which technology is the best solution – don’t start with a robot and try to apply it to everything
- You have to pick the right task to automate. Not all are suitable, either technically or commercially.
Should be an interesting event, afterwards I’ll put a link to a write-up and my presentation here.