Agricultural Robots in Practice: Harvesting, Weeding, and the Labor Gap They're Filling

Agricultural Robots in Practice: Harvesting, Weeding, and the Labor Gap They’re Filling

Walk a lettuce field in California’s Salinas Valley or a berry row in Florida this season and you’re likely to see a machine doing work that, a decade ago, required a crew of stooped-over people with hoes or clippers. Agricultural robots have moved past the demo-day stage. They are weeding row crops with lasers, threading through orchards to carry picked fruit, spraying only the weeds that need it, and — in a few tractor cabs — driving themselves. None of this is happening because farming suddenly became a robotics playground. It’s happening because the people who used to do this work are getting harder to find, and the economics of farm automation finally started to close the gap.

What Are Agricultural Robots Actually Doing on Farms Right Now?

In practice, three categories of machine account for almost everything currently deployed at commercial scale: laser and vision-guided weeders that ride over crop beds and zap unwanted plants without chemicals or tillage; harvest-assist robots that don’t pick fruit themselves but carry it, freeing human pickers from walking loaded trays back to a truck; and a small but growing number of true harvesting robots that use cameras and robotic arms to identify and cut or pluck ripe produce. Layered on top of all three is a wave of autonomy retrofits for tractors and sprayers, letting existing equipment drive itself or target inputs plant-by-plant instead of blanketing a whole field. None of these systems has fully replaced field labor. What they’ve done is take over the most repetitive, back-breaking, or precision-dependent slices of the job, in exactly the crops and operations where finding workers has become the biggest constraint.

Why Is the Labor Shortage Driving Farm Automation?

The labor numbers are not abstract. According to Bureau of Labor Statistics data, the U.S. agricultural workforce fell by roughly 155,000 workers — about 7% — between March and July 2025, a period that coincided with stepped-up immigration enforcement. That decline is notable because it reverses the pattern of the two prior years: agricultural employment had actually grown over the same March-to-July window in both 2023 and 2024. For growers of labor-intensive crops like berries, lettuce, and tree fruit, that kind of swing shows up immediately as unpicked or unweeded acreage, not as a line item to plan around next season.

This is the backdrop every agricultural robotics company points to when explaining why now. It’s also why the pitch for these machines rarely leads with “cheaper than a human.” It leads with “available when a human isn’t.”

How Do Weeding Robots Like Carbon Robotics’ LaserWeeder Actually Work?

Weeding is the single most mature use case in farm automation today, and Carbon Robotics’ LaserWeeder is the clearest example of it reaching real scale. The machine is towed over crop rows, using computer vision to distinguish crop plants from weeds in real time, then fires lasers to destroy the weeds without touching the soil or the crop — no herbicide, no cultivation pass. Since its 2022 launch, the company said in June 2024 that its fleet had eliminated more than 10 billion weeds worldwide, across farms in the US, Canada, Europe, and Australia, in roughly 24 months of operation. Sources differ on the machine’s raw throughput — some report around 200,000 weeds per hour, others cite up to 600,000 — a gap likely explained by different LaserWeeder models or measurement methods rather than a single agreed figure, so treat any specific per-hour number as model-dependent rather than universal.

The company behind it, Carbon Robotics, is led by founder and CEO Paul Mikesell, who had previously co-founded storage companies Isilon Systems and Clustrix before working on deep learning and computer vision infrastructure at Uber — a background that shows up directly in how the LaserWeeder was built as a perception problem first and a farm implement second. Investors have backed that approach at scale: in October 2024 Carbon Robotics closed a $70 million Series D led by BOND, with participation from NVentures (Nvidia’s venture arm) among others, bringing disclosed funding to $157 million at that point. Later reporting suggests the company has raised further, with some sources citing total funding as high as $177 million or even $229.9 million across additional rounds — figures that aren’t consistently corroborated, so the $157 million figure tied to the October 2024 round is the one we can confidently stand behind.

What Does a Harvesting Robot Look Like in the Field Today?

Harvesting is mechanically harder than weeding — ripe produce is fragile, unevenly distributed on the plant, and often needs to be handled gently enough to sell at full price — so progress here looks less like full automation and more like a blend of assistance and selective automation.

On the assistance side, Burro (made by Augean Robotics) is one of the clearest examples. Its autonomous ground vehicles follow pickers through orchards, vineyards, and nurseries, carrying harvested blueberries, table grapes, citrus, or strawberries so workers don’t have to walk loads back and forth themselves. The company closed a $24 million Series B co-led by Catalyst Investors and Translink Capital, with existing backers S2G Ventures, Toyota Ventures, F-Prime Capital, and Cibus Capital also participating — a signal that investors see logistics-within-the-row, not full autonomous picking, as the nearer-term commercial win in specialty crops.

On the fully autonomous picking side, the Agrobot strawberry harvester is the most cited example of a true harvesting robot. It uses robotic arms fitted with color and infrared sensors to identify which berries are ripe and pick only those, and it’s priced at roughly $100,000 per unit — positioned explicitly as an answer to the same U.S. farm labor shortage described above. Reported arm counts for the machine vary by source (some cite 14, others 24), which likely reflects different models or generations rather than one fixed spec, so any specific arm count should be read as approximate.

How Are Autonomous Tractors and Sprayers Changing Row-Crop Work?

Away from specialty crops, the biggest shift in broadacre farming is precision spraying and tractor autonomy, and John Deere has been the company moving fastest on both fronts. Its See & Spray Select system — which uses cameras and machine vision to spray herbicide only where weeds actually are, rather than across an entire field — is now available factory-installed on model year 2026 400 and 600 Series sprayers, fitted with 90-foot, 100-foot, or 120-foot steel booms. It’s also offered as a retrofit precision upgrade kit for model year 2018-and-newer sprayers already equipped with ExactApply and a 120-foot steel boom, which matters because it means the technology isn’t locked into buying an entirely new machine.

On the autonomy side, Deere’s second-generation perception kit — announced at CES 2025 — gives row-crop tractors a 360-degree camera view using 16 cameras, and fits model year 2022-and-newer 9R/9RX tractors as well as model year 2020.5-and-newer 8R/8RX tractors. A separate kit built specifically for orchard tractors swaps that camera-only setup for 7 cameras plus 3 LiDAR sensors, because dense tree canopies block the GPS signal that open-field autonomy leans on — a good illustration of how farm automation has to be engineered differently for a vineyard than for a soybean field, not just scaled down.

Weeding, Harvest-Assist, and Autonomy: How Do the Approaches Compare?

Approach	Example	What it actually does	Crop fit	Key backing/spec
Laser weeding	Carbon Robotics LaserWeeder	Identifies and kills weeds with lasers, no chemicals	Row and specialty crops	$70M Series D (Oct 2024), 10B+ weeds eliminated since 2022
Harvest-assist ground robot	Burro	Carries picked fruit alongside human pickers	Orchards, vineyards, nurseries (berries, grapes, citrus)	$24M Series B, led by Catalyst Investors & Translink Capital
Autonomous harvesting arm	Agrobot	Identifies and picks only ripe fruit	Strawberries	~$100,000 per unit
Precision spraying	John Deere See & Spray Select	Targets herbicide only at detected weeds	Row crops (factory MY2026 sprayers)	90/100/120-ft booms; retrofit kit for MY2018+
Tractor autonomy	John Deere perception kits	Self-driving tractor operation	Open row-crop fields vs. orchards	16-camera row-crop kit; 7-camera + 3-LiDAR orchard kit

Where Does This Fit Alongside Other Physical AI Deployments?

Agriculture shares a pattern with other environments where physical AI is already doing productive work — the same tension between full autonomy and human-assisted automation shows up in warehouse robotics and on factory floors running collaborative robots alongside people. In every one of these settings, the machines that have actually scaled are the ones solving a narrow, well-defined task — carrying a load, spraying a target, cutting a stem — rather than attempting to replace an entire job end to end. Farms happen to be an unusually clear illustration of that principle because the labor gap driving adoption is so directly measurable in workforce statistics.

What’s Actually Holding Broader Adoption Back?

Cost is one obvious constraint — a single Agrobot unit runs about $100,000, and a LaserWeeder fleet or a tractor autonomy retrofit represents real capital outlay for a farm operation with thin margins in a given year. But the bigger constraint may be structural: most current machines are built around specific crop geometries, row spacings, and canopy structures, which is why Deere needed an entirely different sensor suite for orchards than for open fields, and why harvest-assist robots like Burro were designed around specific fruit categories rather than as a universal harvester. Farm automation is advancing crop by crop and task by task rather than arriving as one general-purpose solution, which is consistent with how physical AI is being deployed across other industries — narrow, well-funded, task-specific systems first, generalization later, if at all.

Is the Labor Shortage Actually Solvable by Robots Alone?

Not entirely, at least not yet. The BLS data showing a 155,000-worker decline reflects a scale of disruption that current fleets of weeding robots, harvest-assist vehicles, and autonomous tractors are not sized to fully absorb — these are still relatively small deployments concentrated in specific high-value crops and larger operations that can afford the capital investment. What the current generation of agricultural robots does credibly do is reduce how exposed a farm’s harvest is to a labor shortfall in the specific tasks — weeding, load-carrying, targeted spraying — where they’ve been deployed, and buy operations time while broader labor market and immigration dynamics play out. Investors backing companies like Carbon Robotics and Burro appear to be betting that even partial coverage of that labor gap, across enough crops and acres, adds up to a durable business — a bet that sits inside the same broader wave of capital flowing into physical AI that’s reshaping warehouses, factories, and now fields.

Farming was never going to be the first place people expected to see well-funded, sensor-heavy autonomous machines at work, but the combination of a genuinely shrinking labor pool and increasingly capable computer vision has made it one of the more concrete proving grounds for the broader physical AI applications already running in warehouses and factories. For a fuller sense of how terms like autonomy levels or perception stacks apply across these deployments, the glossary is a useful reference point as the vocabulary from robotics keeps showing up in farm trade press. What’s clear from the machines already at work in fields today — the LaserWeeder’s laser count, Burro’s orchard runs, Deere’s dual sensor suites for open fields versus canopies — is that this is incremental, crop-specific progress rather than a single robotic solution to farm labor, and the pace of that progress is now being set as much by workforce numbers as by what the technology itself can do.