Agriculture is a field where robotics holds particular promise. Agricultural work is often simple but grueling, and the agriculture and horticulture industries are finding it difficult to meet their labor requirements. A startup robotics company called Harvest Automation is developing robots designed specifically for agricultural work. In an interview with Sander Olson, roboticist and Harvest CTO Joe Jones describes behavior-based robotics, the future of robots, and how robotics could become a transformative force in the agriculture and horticulture industries.
Question: I first interviewed you 10 years ago. How has the field of robotics evolved since then?
The robotics field is considerably larger than it was a decade ago, and there are now several successful commercial companies providing robotic equipment. The robotics industry is more focused than in 2001. Back then, many researchers envisioned creating general-purpose robots. In 2011, we are thinking more in terms of purpose built robots that accomplish a particular task, rather than general purpose robots. I see that as being a positive development.
Question: You originally worked for iRobot on the Roomba robot. How similar are these agrarian robots to Roomba?
Roomba is different in many ways. Roomba was only designed to operate indoors, on flat surfaces, and it had no idea of where it was. It wasn’t waterproof, and couldn’t affect its environment other than cleaning the floor. By contrast, our Harvest robots have to know where they are, they have to grip and manipulate pots, and they have to be able to operate in a wide variety of terrain and field conditions.
Question: These harvest robots are battery powered. Did you consider using gasoline engines?
We considered it, but ultimately decided against it. These robots are designed for both outdoor and indoor operation. To develop a propane engine for indoor use would have been costly and unnecessary, so we went with batteries. Batteries have improved, and we use rechargeable lithium-ion batteries that should give us a 4 hour runtime. Users will simply swap out the depleted batteries and put in fresh ones.
Question: How do these robots navigate?
We wanted to develop a simple sensor system that was robust and low cost. GPS systems that can navigate to within an inch or two are quite expensive, so we developed an alternate approach that involves using boundary markers. This boundary marker is made of a reflective material that robots easily recognize. The robots pick up a plant, move toward the boundary marker, and then follows the boundary marker to determine where to put the plants. The robots only need to position plants a few feet away from the boundary, this limits the amount of dead reckoning robots must do and that, in turn, limits the build up of positioning error. This method is much simpler and less expensive than alternate methods while providing the level of performance and accuracy that we need.
Question: So these robots lack cameras?
They lack cameras, because cameras are poorly suited to working in sunny areas and our robots need to work in open field environments. Cameras don’t work well with both very dark objects and bright reflections in the same scene, so we went with a simpler and more robust system. But, with further development we are looking into adding cameras for future models to enhance the functionality of the robots.
Question: Wouldn’t radar be useful?
We considered radar as a ranging system, but many of our pots are only 4-8 inches in diameter. The radar “beam” is too broad to work well with narrow objects—it’s lateral resolution wasn’t sufficient.
Question: Each of your robots can discern other robots. Why is this important?
Teammate detection is important because we need the robots to operate at a high level of performance and throughput to meet the needs of growers. Each robot checks to see if there is another robot in the space it wants to traverse. If it discerns a teammate, it waits for the teammate to pass. Currently, the robots, can differentiate between pots, other robots, and obstacles.
Question: And they manipulate pots using a gripper?
Yes, our manipulator has two degrees of freedom, with a simple gripper at one end, for lifting and depositing plants. It is capable of lifting up to 22 pounds. Our robots will be able to move 1-5 gallon size containers which represent the majority of the plants produced by the nursery & greenhouse sector.
Question: What sort of AI do these robots employ?
We use behavior-based programming, and these robots are capable of performing a single task intelligently. Our robots are capable, for instance, of “marking” objects so that it won’t attempt to pick up the same wrong object (for example, a sprinkler standpipe) twice. They also can “change their mind” about what to do next based on ongoing changes in the environment.
Question: Could these robots be used to spray plants, or to pick fruit from trees?
We are already working on concepts for robots that spray pesticides and herbicides, as well as performing other critical production tasks, but a commercial version of these machines is several years in the future. Trees are out of reach, no pun intended. Picking fruit is a surprisingly difficult problem.
Question: Potentially how large is this market?
Although I can’t provide exact figures, our research has concluded that the Agriculture industry annually spends in excess of $20B on low productivity manual labor that could be addressed by agrarian robots based on our platform. Anyone who has tried to move heavy plants in 90+ degree weather knows that it is grueling work. So it is no surprise that the horticulture industry has great difficulty finding human workers willing to do it. Every year the cost of labor will increase, and every year our robots will become more capable.
Question: How much will these robots cost? Will there be different versions of the robot available?
We are targeting a pricing level that provides the grower with a 12 to 24 month payback on the robots. Finding legal manual labor is a difficult challenge for growers—the costs and uncertainty increase every year. Growers value our robots because they provide a dependable, low-cost solution to the problem of worker scarcity. We have received numerous advanced deposits from growers who want our robots. This is an indication of how much interest there is in our product. We see our current product as the first robot in a line of products that we will develop for the Agriculture industry.
Question: How difficult would it be to modify these robots for various tasks?
It should be fairly easy. With Roomba, there wasn’t much that we could do with it outside of cleaning the floor. By contrast, this technology is very adaptable to other tasks that agriculturalists require. The capabilities of our robots will rapidly increase, and we envision them being used for all manner of agricultural tasks that humans don’t like doing.
Question: The military and healthcare fields both hold enormous potential for the robotics industry. Why not enter those fields?
When we founded the company back in 2008, it wasn’t pre-determined that our focus would be on providing robotic equipment for Agriculture industry. We initially looked at several potential markets including possible military and healthcare applications, but concluded that the low productivity sectors of the Agriculture industry involving highly repetitive, physical work represented the greatest opportunity for commercial robots in terms of a well-defined need that could be addressed with existing technology.
Question: The Apple Siri software is enormously popular. Why not add voice-recognition capabilities to your robots?
Voice recognition is fun but unnecessary. The primary objective with robotics is getting the robot to perform the task that you require it to perform. Having a user interface that makes it easy to tell the robot to do the job is the subsequent step. We had the opportunity to field test our user interface at grower operations this summer and the response we heard back was overwhelmingly positive. Perhaps this voice-recognition capability will be standard on robots a decade from now, but this is not high value for our product and not my area of specialty.
Question: What will be the first “killer app” for robotics?
There have already been several successful robot products including but not limited to Roomba and the robots that are provided by Kiva Systems. The idea of a “killer app” may be more suited to the software industry, but we believe that Harvest is developing a “game changer” and this is just the first in a line of products that we envision will revolutionize a big part of the Agriculture industry.
Question: What would you recommend to someone thinking of founding a robotics startup company?
The key is to focus on a specific application where there is a defined need. There are quite a few manual operations that could profitably be done by robots today, without any technological breakthroughs. The problem is that the people doing these chores aren’t aware of robotics, and the people doing robotics research aren’t aware of these tasks. Tremendous opportunities exist for individuals who can connect these two domains.
Question: Will we see the first general-purpose robots emerge within ten years?
Ten years is too soon for general-purpose robots, but there could be dramatic progress within the next decade. I predict that by 2021 we will see all manner of robots that are able to effectively and cost-effectively be able to perform a wide variety of specific tasks. The growth in robotics may not be as fast as the growth in the computer industry, but it will still be quite rapid and the value creation is significant.