The one thing stopping robots from working inside a polytunnel
Drive a farm robot into a polytunnel and something quietly breaks. The satellite signal that was guiding it — fixing its position to within a metre or two across open ground — scatters off the metal frame, degrades under the polythene canopy, and becomes too unreliable to act on. The robot either stops, drifts, or requires a human to intervene. For most commercially available systems, that is not an edge case to be patched: it is a fundamental constraint of how they navigate.
Polytunnel and protected-environment growing is a substantial part of Lincolnshire's horticultural output. The county produces 30% of the UK's vegetables, much of it in precisely the kind of low-clearance, high-structure growing environment that defeats standard GPS waypoint systems. The same problem appears in vine rows, under dense crop canopy, and anywhere steel infrastructure is close at hand.
Conventional robot deployments have typically worked around this by requiring pre-deployment setup: a technician walks or drives the boundaries, the system maps the area, and the robot runs fixed routes. For arable fields that change little from season to season, that is manageable. For seasonal polytunnel operations — where layouts shift, crops change, and set-up time is itself a cost — it is often impractical enough to rule the technology out entirely.
The result is that farm robots have remained largely peripheral in protected horticulture despite nearly a decade of serious investment in agricultural autonomy.
What JABAS.AI actually does differently
Rather than waiting for a satellite signal that may never arrive inside a polytunnel, JABAS.AI's perception stack builds its own picture of the environment as the robot moves. Lidar — a sensor that fires rapid pulses of laser light and measures how they bounce back — gives the system a continuous three-dimensional read of its surroundings. Computer vision layers on top, identifying crop rows, obstacles, and the edges of working areas. Together they generate a map on the fly, updating in real time rather than relying on a fixed chart laid down before the season started.
That distinction matters practically. Because no pre-deployment walk-through is required, operators do not need a technician to re-map a site each time a layout changes between cropping cycles — a meaningful saving in seasonal growing environments where conditions rarely stay static.
The platform also tracks where human workers are within the fleet's operational area, adjusting routing dynamically so that machines and people share a space rather than being segregated from one another. Dynamic boundary detection performs a similar function at the field edge: the system recognises limits as conditions change, rather than stopping at a hard-coded perimeter.
Commercially, the most significant aspect may be what JABAS.AI is not. It does not sell robots. Instead, the software acts as a coordination layer — a middleware that can sit on top of machines from different manufacturers, allowing a mixed fleet to work together on tasks ranging from weeding to logistics. Described by the company as 'autonomy-as-a-service', this model means a grower's existing hardware can gain fleet-level intelligence without replacing it. That is a meaningfully different proposition from buying a proprietary machine, and it is currently being put to the test with six commercial farming operators.
Why Lincolnshire is an unusually high-stakes test bed
The scale implied by that polytunnel GPS problem is easier to grasp once Lincolnshire's food production numbers are set out fully. Beyond the vegetable figure touched on earlier, the county accounts for 12% of England's entire food output and holds more Grade 1 agricultural land — the highest quality classification — than any other Local Enterprise Partnership area in the country. Agricultural output topped £2 billion in 2019.
Those numbers acquire a human dimension when set against employment. Food and farming account for 24% of all jobs across Greater Lincolnshire, close to double the national share of 13%. A technology that shifts farm productivity even modestly does not stay contained at the farm gate; it moves through a workforce whose reliance on the sector is unusual even by rural standards.
Post-Brexit restrictions on seasonal migrant labour sharpened that pressure considerably. The most visible large-capital response in the county has been Dyson Farming's 26-acre fully robotic strawberry glasshouse — a facility that represents one answer to the question of how labour-intensive horticulture continues when the seasonal workforce shrinks. It is not a universal model; the capital requirements alone put it beyond most growers.
The environments the bulk of those growers work in — polytunnels, vine rows, crop canopy — are exactly those where GPS navigation fails. That is not incidental to JABAS.AI's development: its trials are running in precisely these settings because they are where autonomous coordination has to prove itself, not simply where conditions happen to be convenient.
Reframing the robot from threat to load-bearer
One in six working hours on a horticultural holding may be spent not harvesting but carrying — moving produce between picking rows and collection points. That 15–20% transit-time figure, cited in the research underpinning JABAS.AI, is the specific gap the platform's fleet logistics aim to close. It is also the number that reframes the labour argument.
The question any farm worker has reason to ask is whether a robot fleet reduces the number of people employed or reduces the burden on those who remain. Those are different propositions. When the task an autonomous vehicle takes over is repetitive transit rather than skilled harvest judgment, what nominally disappears is the carrying, not the picking. That is a meaningful distinction — and in a labour market already under pressure from post-Brexit workforce contraction, it is the more persuasive pitch for growers looking to extend capacity rather than simply reduce headcount.
Carrying is, in one sense, the part of the working day that demands the least from a skilled picker's hands and attention. Recovering those hours for productive harvest work is a different pitch from replacing a worker entirely. Whether it holds as a genuinely stable outcome — or whether fleet automation eventually affects total employment on participating farms — has not been settled by independent local analysis. What the six operators currently running JABAS.AI report, when those field seasons conclude, will carry more weight than any advance framing.
The institution behind the spin-out
JABAS.AI is not a speculative start-up pitching a prototype. It is the fifth spin-out from Ceres Agri-Tech — a partnership between the universities of Lincoln, Cambridge, and East Anglia, funded by Research England and EPSRC — and it emerges from nearly a decade of research at Lincoln's Centre for Autonomous Systems and the Lincoln Institute for Agri-food Technology.
The research centre behind it, Lincoln Agri-Robotics (LAR), was named the world's first global centre of excellence in agricultural robotics in the UK Government's Innovation Strategy in July 2021. In February 2024, it received the Queen's Anniversary Prize for innovation in agri-food technology — an independent peer review of research quality, not an industry award. Since 2017, the underlying work on long-term robot autonomy has attracted £4 million in Innovate UK and direct industry funding; foundational papers have appeared in Nature.
Professor Marc Hanheide, LAR's director and JABAS.AI's CTO, brings that lineage directly into the company. The wider Ceres portfolio — which also includes Agaricus Robotics, focused on mushroom harvesting, and FruitCast, an AI fruit-forecasting tool — has collectively created 34 high-value rural jobs across the region. That figure puts a human number to what 'spin-out portfolio' otherwise means in the abstract.
What still needs proving before this changes how Lincolnshire farms
Six trial sites across polytunnels, vineyards, and open fields is a meaningful start. It is not yet a business case.
No commercially published performance data from those six operators has entered the public domain. Growers who want to model return on investment — the cost-per-hectare calculation that would let a farm manager weigh a software subscription against hiring an additional seasonal worker — cannot yet do that from available evidence. Field seasons have to run to completion before any honest figure is possible.
The terrain question is also open. JABAS.AI's validated environments, as publicly documented, are polytunnels and vineyards. Whether lidar and computer vision handle heavy clay, drainage channels, and large-machinery co-navigation routine across Lincolnshire's arable land has not been established in public sources. Those are not exotic conditions in this county; they are an average Tuesday in the fens.
There is also the hardware dependency. As a software layer, JABAS.AI needs robot platforms to integrate with. Growers who have not yet invested in autonomous hardware face two adoption hurdles, not one.
Perhaps the sharpest local question is scale. Most Lincolnshire farms are not Dyson-scale operations. What this technology costs, and what it returns, on a 60-acre family holding growing brassicas rather than a 26-acre glasshouse designed from the outset for robots — that is the number the next harvest season actually needs to produce.