Red harvest: robot farmers find sweet spot for strawberries

Dogtooth Technologies, a start-up company based near Cambridge, is developing autonomous robots designed to harvest soft fruits such as strawberries.

The company tested its first prototype robot in the field in the summer of 2016, and is already working on an improved version. Five of these second generation robots are due to be operating on a farm this summer.

Now, in an EPSRC-funded project, known as Vesca, the company is working with researchers at the National Institute of Agricultural Botany in Cambridge to develop advanced computer vision and motion planning systems, to enable the robots to more efficiently locate fruit and determine their quality and suitability for picking.

The team will do this by improving the robots' machine learning algorithms. This will enable them to make better decisions about how to approach target fruit, and whether to pick them or not. This should result in an increased pick rate of saleable fruit, whilst minimising waste.

The robots are each equipped with several cameras, said Ed Herbert, founder and chief operating officer of Dogtooth Technologies. “You have got these robots moving up and down the fields with several cameras on, and they’ll be imaging the crop in minute detail,” he said.

“We have got that linked up with high precision GPS coordinates, which allows us to look at the trajectory of each plant and each strawberry,” he said.

By tracking the trajectory of a given plant or section of the farm, it allows farmers to quickly identify an area of the farm that is more or less productive, or where there has been a sudden drop-off in activity, said Herbert.

“You can look for signs of disease or pest, and that means that you can intervene earlier, reducing the amount of chemicals you need to use, and increasing your yield,” he said.

The project is aiming to build a system that has approximately the same performance as a human strawberry picker, but at a cost of £12,000 per robot, when produced at scale.

Strawberry harvesting is manually intensive, and increasingly vulnerable to fluctuations in the labour market.

“But picking is almost an excuse to get cameras in the field,” said Herbert. “The real benefit is not in replacing or displacing human labour, it’s in all the added information you can get with these really rich datasets we will be collecting.”