--As told to Leigh Buchanan.

Some jobs are so mind-numbing and backbreaking that no one with other options wants to do them. Cue the robots. Harvest Automation, a 31-employee startup in Billerica, Massachusetts, designs robots that arrange pots in nurseries and greenhouses. It's a monotonous task, and the associated labor represents 40 percent of growers' costs in a $17 billion market. So far, Harvest Automation has about $25 million in funding and 29 customers, most of which start out with four of the $30,000 bots. Soon the robots may perform other jobs, too, including warehousing and manufacturing. Co-founder and CTO Joseph Jones recently spoke with Inc. about how the company started.

Flying the iRobot nest

The other three founders and I met at iRobot. Three of us were on the team that developed the robotic vacuum cleaner Roomba. Roomba's success was gratifying but also frustrating. By 2006, iRobot had gone public, and there were fewer opportunities to work on projects outside the consumer-products or military realm. Our COO, Charlie Grinnell, and I started talking about robots that would be both innovative and commercial. Charlie said that since anything we came up with would belong to iRobot, we should strike out on our own. We set ourselves up in his house and spent the next few months deciding what to build.

Our three laws of robotics

We didn't have a product at first, but we did have some principles. First, too many robots are created with multiple features so that they can be deployed in multiple situations. Those general-use robots can do lots of things but excel at none. We would make a robot that could do one job extremely well. Second, most work is done in systems, often with multiple handoffs. Ideally, you want to divide these systems into tasks that are best performed by robots and tasks that are better for people. That meant the robot needed to be safe enough to work alongside humans. Third, the robot would have to be mobile and able to operate in the world we inhabit rather than in a specially engineered environment.

Room to grow

We attended trade shows in search of ideas. At one for horticultural growers, somebody told us about an industrywide problem: spacing out pots. Roughly two billion ornamental plants are sold every year. Those plants start out as seedlings in pots that are generally packed close together to maximize space. As they grow, they need more room and the pots must be separated, sometimes over and over as the plants get bigger. People--typically migrant workers--have to bend down and move those pots, keeping the spaces between them consistent. They do this in all weather conditions, risking repetitive stress injuries. The optimal jobs for robots are dirty, dull, and dangerous. This one qualified on all three fronts.

Testing the waters

Before developing the robot, we talked with potential customers. They were eager to get something like this, because they have trouble finding people willing to do the job.

We created a partner program so we'd have a better understanding of how the robots fit into the workflow. Growers could pay $20,000 to participate, which would go toward the pur­chase price of a robot. About a dozen companies signed up, and we tested the robots onsite at about half of them.

As a result of those tests, we made several changes. We repeatedly modified the software and hardware until our robots could negotiate bumps, ruts, and gravel.

Robot farmers

We think our robots could be of use on farms, too. We started a pilot program in California to study the potential for growing specialty food crops--such as strawberries and tomatoes--in pots and using robots to space them out. That would make it faster and easier for farmers to go organic. They would no longer be restricted by the soil on the ground but could instead fill pots with the requisite growing media. The robots couldn't harvest the food, but they could carry the pots in to workers to do the picking, which would prevent injuries. That process would also allow farmers to plant more crops, because robots don't need space to walk between plants. Farmers could fit 30 percent more strawberries on an acre of land, according to our calculations.

Our robots have the potential to help with pests, too. Instead of having to blast a whole field with chemicals, our robots could help farmers find infested areas. We've filed for a patent on a robot that can examine and detect problems in individual plants, allowing farmers to narrowly target their responses. That should reduce the presence of chemicals in the food supply.

Improving--not stealing--jobs

We were naturally concerned about accusations that our robots would steal jobs. Charlie talked to customers about that early on. They said, "Don't worry. We're not going to fire anybody." Currently, growers have a shortage of workers, so they plan to keep them on and give them higher-value tasks. And the workers we are training tell us they would much rather supervise robots than move pots around by hand.

A better, cheaper bot

One of the challenges now is making the robots more reliable. These devices drive on potentially nasty terrain--usually more than 10 miles a day. So there is a lot of bouncing and shaking. The only way to address that is to keep testing them ourselves and gathering data from customer sites. If something fails, we analyze why and, if necessary, redesign the robot so it doesn't happen again.

We also have to make them less expensive. When you make the first one of something, you typically overdesign. Then you can assess what is and isn't important, and redesign it at lower cost.

As for competition, there are certainly other options in materials handling, which is the next area we are targeting. But we have an edge, because we build things to operate outdoors, where you don't control the environment. Plus, to add another company's robot to a warehouse, you'd have to take out all the items in the warehouse to install the system the supplier created. Our robots are autonomous enough to deal with what's already in there.

The Reject Pile

Separating pots in nurseries was one of 15 competing ideas that Harvest Automation's founders considered. Among those that didn't make the cut:

Moodroid
A farm-dwelling robot to pick up cowpies and deliver them to a methane digester for conversion into energy

Doodroid
A house-dwelling robot to pick up dog feces from lawns

Norobot
A ship-dwelling robot to patrol the passageways and test for signs of the norovirus--the vomit-inducing bane of the cruise industry

Aquacine
A water-dwelling robot that would travel along the bottom of a pool to film swimmers in training as they do their laps

From the February 2015 issue of Inc. magazine