SlugBot Trials: The journey and next steps

By Olivia Cooper

Individual recognition of target species in a field and spot treatment by a robot, without chemicals, is a giant leap closer to becoming a commercial reality thanks to the SlugBot project. Olivia Cooper finds out more.

What started out as a Nuffield Farming scholarship into the future of slug control in 2018 rapidly snowballed into a ground-breaking research project which has incredible implications for the world of machine learning, AI and autonomous pest control. And it’s been delivered on-time, in-spec and on-budget, despite taking place during the Covid lockdown. So how did it come about?

“While on my Nuffield travels, I realised there was a massive gap in the market to improve monitoring and biological control of slugs,” explains Dr Jenna Ross, international business development manager at UK Government funded AgriTech centre, Crop Health and Protection (CHAP). “Most farmers either monitor by hand using traps, or not at all. Blanket application of metaldehyde led to significant impact on non-target organisms and environmental damage and thus its subsequent ban. And I realised technology could play an important role in filling that gap.”

A big advocate for biological control, Dr Ross realised that the use of nematodes for slug control would be prohibitively expensive unless precisely applied – so she set about finding out how to do it. With the help of Innovate UK funding, she was able to utilise the CHAP facilities at Rothamsted Research to identify unique spectral signatures for slugs.

The first step was to collect lots of slugs to test the multi-spectral imaging capability. They chose to target two pestiferous species, the Grey Field slugs and Spanish slug. Just one problem – this was April 2020 and nobody was allowed to go out. “So we opened a citizen science project and asked people to send in their slugs,” explains Dr Ross. Having identified that they do indeed have a unique spectral signature, she then started to breed the slugs in order to have sufficient specimens and images to train up the AI model.

SlugBot - the autonomous monitoring and treatment system.

And this is where SRC comes in; the Dick robot had already been built and the Wilma computer had considerable AI capabilities, but could they be trained to identify and spot treat slugs? Andy Hall, robotics product manager at SRC, was running the prototypes and looking at non-chemical weeding, so there was already a decent crossover, he says.

Originally, SRC was going to outsource the AI work to specialists, but Innovate UK limitations meant it ended up recruiting and creating its own AI team. “That’s probably been the most valuable thing we’ve got from the project, although it did cause a few headaches at the time,” says Mr Hall.

In order to get photos of slugs in their natural environment, CHAP planted a polytunnel with oilseed rape and wheat, and SRC created a slug hotel at its headquarters. “We took lots of images in the lab, then in the glasshouse, and finally in the field, so we had a mix of controlled, semi-controlled and real conditions,” explains Dr Ross.

“It was really quite funny,” adds Mr Hall. “We took the slugs along and placed them in the crops, then drove the robots around to take photos of them. We had to do it in the dark, so we were trying not to crash into the polytunnel!”

Tom took the photos and uploaded them to Wilma to be processed and identified. One more problem though: Slugs move. “That was a fairly fundamental thing to overcome,” he explains. We developed Edge AI, which is specially trained to identify slugs and sits on the Tom robot, so it can immediately trigger the spray application.”

The 1m boom, which was adapted from a conventional spraying system, sprays an area of 20x20cm, and developing it really started SRC on the road to spot spraying. “As a company we’ve learnt a lot. We’re using existing nozzles in a different way, controlling the pressure and flow rate as it’s travelling so much more slowly than in a conventional sprayer.”

The spray contains nematodes, which are a natural parasite of slugs. Once in the field, the nematodes actively search out slugs and infect via natural openings. The nematodes then start to reproduce causing a characteristic swelling of the mantel area. Death of the slug usually occurs between four and 21 days after infection, and once the food source is depleted, the nematodes move back into the soil to search out new slug hosts.

Having proven the concept, the team then took to a farm in Devon to see how it worked in the field. “It was really important that we had a farmer on board, and James Lee – who was a previous AHDB Monitor Farmer – is very interested in R&D,” says Dr Ross. Once again though, the activity had to take place at night when slugs are most active. “It was like War of the Worlds,” laughs Mr Hall. “We had these robots going round in the dark with lights flashing – it was quite something.”

So what’s next for SlugBot? Dr Ross is currently looking for follow-on funding to continue improving the data package and spraying abilities. “It’s been a great project and generated a huge amount of interest,” she says. “But there’s a long way to go. We need more data and more images. This is just the tip of the iceberg.”

This project has provided SRC with new opportunities to progress, including the modular systems developed with the AI.

The most exciting part for her has been discovering that multi-spectral imaging can be used to identify a pest in the field. “That means you can swap out the slug to identify other pests, diseases, weeds or plants, which opens up so many opportunities for both remote sensing and the use of biologicals.”

And the project has opened up new opportunities at SRC, too. “The modular systems we’ve developed with the AI have become a core part of the company,” explains Mr Hall. He has particularly enjoyed the variety of developing such cutting-edge technology. “One minute you’re dealing with GPS and satellites, the next minute it’s electronic control and mechanical equipment – we’ve learned so much.

“The ability for a robot to find something, identify it and take action has come about as a direct result of the project, too. We’re now looking at using the technology in other areas – in two years I think we’ll have a precise spot sprayer as a result.”