Category: Sensors

Is the the end of the ag robot swarm ?

An academic agtech dream has died

I am personally sad that Small Robot Company has gone into liquidation. Personally sad because I invested in it but also because it is probably the end of an academic dream that friends and colleagues believed in passionately. The dream failed to make it to market for a number of practical reasons.

  • Finance
  • Minimum Tillage
  • Weeding
  • Harvesting

Finance. Interviewed on BBC Radio 4 Farming Today the founder of Small Robot Company blamed the Agtech financing climate in the UK and I would agree from my experience starting eCow, Milkalyser and VirtualVet there are few investors or funds willing to take long term high risk investments in UK agtech. Investors mostly want companies with proven products and sales. Any company developing hardware that needs to be tested and developed over a number of seasons is going to struggle. Even if the product works and sells the cost of scaling is immense. Investors love software start ups because testing is usually quite quick and upgrades can be cheaply sent out over air. There is plenty of evidence that the Silicon Valley model of funding does not work well in farm machinery. These are not fast moving consumer goods bought as fashion items.

Minimum Tillage. For years the concept of reducing the soil compaction associated with heavy machinery has proposed that fleets of light weight robots could replace the heavy machines needed for field operations. The Tom robot mapped pests and then returned for spot spraying. The concept was that the major time cost for human labour could be replaced by a few machines taken out to the fields that would quietly work away. However, I think the real problem of compaction came from the previous fashion for deep tillage which requires huge motive power and thus heavy machines. Minimum or even no till requires a completely different approach with use of tramlines and accurate GPS. If spraying is to done then 24 m passes on tramlines and lighter machines that could be automated rather reduce the argument for the swarm of small robots.

Weeding As we move away from spraying weeds towards steered blades, electrocution and lasers we reintroduce the need for power on the implement. My abiding memory of the wonderful laser weeders working rocket beds is the sound of the diesel generators needed to power the servers and lasers. These will not be powered by small machines in the near future. Blade weeders need traction and good electrical supply too for the imaging computers.

Harvesting The biggest machines we always see are always harvesters after all we planted kilograms of seed and expect to harvest tonnes of crop. Cereals and forage crops get harvested when soils are dry in summer when compaction is lower risk but the high value vegetable crops with big labour demands have to be brought in over often soaking soils so again we need power.

In my view the future of field robotics is to develop systems that allow farmers the flexibility of use of tractors and implements with automation tools that for routine field operations do not require a driver. I am working on remote control systems that can be used from off the seat but stop safely if humans are in danger around the machine. Only the largest farms can justify machines that can do limited operations.

Why Artificial Intelligence (AI) will not create robot tractors

Twenty years of incredible development have failed to replace humans as drivers.

Twenty years ago DARPA who sponsor advanced research in the USA ran a competition to develop a vehicle to save lives of the US military. This was hoped to lead to autonomous vehicles. This excellent book takes us through the exciting developments of the past 20 years of sensors and systems but we still have no autonomous vehicles on public roads. Exciting, and initially tech driven, the competitions in the Mojave desert are particularly well described. The first failed to find a vehicle that could complete the course but two years later several vehicles managed the 150 miles of rugged terrain. The action then moves to the even more dynamic environment of urban driving and the businesses that were created and failed to deliver a safe system to replace the human skill of driving. There is so much in this book of interest to engineers, drivers, business people and investors still looking for the secret sauce that will bring sensory intelligence to robots. Things that humans do with ease are difficult for robots and vice versa. This is Moravecs paradox explored with billions of dollars of investment money from Google, Uber, Ford, GM et al. and still there are no autonomous vehicles regularly on the road.

For me, with a Defra funded innovation project to develop safe working systems to stop tractors killing humans and animals this is a key text. The book maps out the development process that failed to constrain the operational design domain and demonstrates the hubris of some inventors one of whom ended up in prison for IP theft. The desire for a complete system prevented the development of the aftermarket of conversions of existing vehicles that could evolve autonomy. Whilst deep learning techniques of showing many tagged pictures of cats to a computer can teach it to recognise a cat, this has proved insufficient to learn the complexities of navigating a road without accidents.

For me this is welcome. I am disappointed with the displays of robotic tractor systems and there are many now appearing, they seem to have to have a narrow focus and few farms are big enough for a single purpose machine.

I want to keep the flexibility of the existing tractor fleet and provide a conversion kit that allows them to be used in many tasks, anything from feeding animals on a frosty morning or carrying tools and posts while fencing. I have defined a few problems where humans work with tractors and spend a lot of time jumping on an off and driving short distances.

#tractor #robot #farming #defra #innovateUK #horticulture #robothayes

Invasion of the non-Hollywood Robots

A great show of the latest field robots in 2022

In a spirit of curiosity and in need of an adventure after the confinements of the last two years I took flight to California for a bit of October sunshine to visit FIRAUSA2022 to see the latest display of Californian Agricultural Robotics. Last year I managed to navigate Covid restrictions to visit FIRA2021 in Toulouse and my blog report of that meeting is here.

FIRA USA 2022 was held in Fresno a modern city of half a million at the Southern end of Central Valley region of California which is one of the great growing areas of the world where with irrigation water from the mountains pretty much anything can be grown. At the airport one is greeted with Welcome to Fresno, agricultural capital of the world.

The meeting was enjoyed by 1000 + delegates in a three day mix of R&D results, commercial interactions and most interestingly field demonstrations in salad crops and vineyards.

Figure 1: The design of full autonomous tractors completely reshapes the layout. They are basically a frame with wheels at each corner, a power source, sensors and a computer. This Naio machine has front mid and rear tool bars on hydraulic hitches.


Being California there was huge enthusiasm to invest in the new wave of technology based on autonomous robotics, image processing and precision navigation. I think the investment crowd may have been a little disappointed as various company founders expressed views that suggested company growth should be organic and that VC money was not always appropriate as rapid scaling of agritechnology and early exits were unlikely.

The main interest was in systems that address the key problem in horticulture and viticulture of labour shortage. Even Mexican growers present were complaining that there is a labour shortage there. So weeding by hoes, spot spraying or lasers were the most viewed systems along with sub systems suppliers for navigation, wifi charging or image processing etc,

Most of the blade weeders were relatively simple, agitating a blade to swipe a weed identified by image processing as not the crop. One company FarmWise had build a full autonomous vehicle to weed which begs the question why not just mount it on a tractor and apply navigation and autoturn controls such as those offered by several manufacturers now.

Figure 2: interplant and interrow weeding just has to identify a plant as not the crop to then chop its roots with a blade. The red circles are weeds left to dry out in the sunshine of California


To me all the blade weeders looked primitive in comparison to the Garford Robocrop invented by Nick Tillet and Tony Hague 20 years ago and this was also pointed out by one of the scientific speakers.

More exciting was the laser weeder that could operate in beds of very small seedlings as it could burn out out weeds with pinpoint accuracy.

There were a lot of interesting panel discussions and science presentations and one I caught said that blade weeders were over 90% effective weeding between lettuce plants in rows although timing is critical. In Arugula (Rocket in UK) where plants are grown in beds timing is even more critical and there a laser weeder came out very well. You can see it working in the video . I walked down the bed after the machine finished and saw only two weeds in 100 metres. Both were emerging contiguous to the Rocket seedling which would need a finger weeding if a human was doing it.

There was a lot of attention on viticulture robots mostly for spraying which is obviously a major operation. See the video for a view of some of them. My favourite machine is the Swarmfarm system which was not able to make the journey from Australia where a number of them are busy. They are basically a metal frame, a diesel power source, navigation and autosteer. The video demonstration is well worth watching. I spoke to Andrew Bates the founder and he said he has a couple coming to UK soon.

There was considerable discussion about whether legislation will hamper the deployment of autonomous robots on farms. No-one wants a farm worker or child hurt by a robot although scaring off ramblers and crop circlers might be acceptable.

The Californian equivalent of HSE have recently judged that fully autonomous tractors must be supervised by a human and this has dealt a major blow to progress on autonomy.

This maybe a hurdle we need to jump in the UK fairly soon as autonomous tractors are working at a number of farms in test conditions.