A new omnidirectional, vision-based localization and mapping system is being developed at the University.

Aerospace engineering assistant professor Soon-Jo Chung, electrical and computer engineering professor Seth Hutchinson and mechanical science and engineering Ph.D. candidate Junho Yang are working on a navigation system that will allow objects like lawn mowers to safely and autonomously function.

They have partnered with John Deere in hopes of bringing this navigation system technology into the company’s robotic lawn mowers and introducing them to the U.S. market.

The Daily Illini spoke with Chung about how the technology they are developing works.

The Daily Illini: How did the idea for a robotic lawn mower come about?

Soon-Jo Chung: The development of a robotic lawn mower is not really the focus of our project. John Deere is already selling robotic mowers in the marketplace. It’s just that they have not introduced their robotic mowers to the U.S. market; they are selling them in Europe.

What we are doing here is to further improve the autonomy and performance, or maybe convenience, of this robotic mower, so potentially it can be introduced to the U.S. market.

DI: What exactly does your research project look like?

SJC: Basically, it’s a work between myself and my colleague Seth Hutchinson in electrical and computer engineering. We are all part of faculty at the Coordinated Science Lab; it’s an inter-disciplinary research unit in the College of Engineering. Our project is to look into using some computer vision and some autonomous recursions to replace the existing navigation system of this John Deere robotic mower.

The concept that they are using is that they use these boundary wires that have to be embedded or installed prior to first use, so that they contain this mower within the boundary. It’s inconvenient, and if you move to another area or another home, then you have to reinstall these wires.

But if you actually recall the popular vacuum robots from iRobot, Roomba, and how they actually can be contained in such a region … For example it might fall off the stairs, so there is an infrared system, infrared-based beacon. So you can buy additional beacons that basically are installing virtual walls if you have to use the Roomba.

It’s like that, so a boundary wires system is like you are constructing an actual virtual fence. It’s actually somewhat similar to these invisible fences you have to use for your dog.

Bottom line, the current John Deere robotic mower, or any other robotic mower available in the marketplace, uses this kind of virtual invisible fence system where they have to install wires to declare or determine the boundary of your lawn.

What we are trying to do is what if we can actually get rid of that so that it can be more flexible? We are going to use our vision, this kind of computer vision, so we are going to mount these omnidirectional cameras. We need a computer to process these visions … to accurately predict and estimate the location of the mower and establish some boundary that it can actually find.

The system we have to add to this mower includes omnidirectional cameras. Basically, this is a 360-degree field of view. Typically, you have a certain field of view limitation; you can only see certain angles from your camera. This one can look at whole views surrounding the location.

DI: The robotic lawn mower and this technology that you are developing, does it have some sort of safety features? Can it sense if there are children around or dogs or if there are obstacles?

SJC: Yes, computers just have a wire-based system. The only thing you can do is maybe it can be contained in your lawn, right? But what if your children are running around and there are many paths? So, we are using a vision-based system to detect the boundaries of your landmarks, like trees or doors and windows. This actually is composed of two-step processes.

First, the mower can follow the boundary of your lawn. While it’s doing so, the computer vision-based autonomous navigation system of this robotic mower detects and finds and registers all those features found on the camera, like windows or trees or doors, so that it can register those images and features along with the actual position of this mower.

What happens is that next time you turn on the switch … The first step is called the training mode; you just train the actual boundary of your lawn. The next time, it can automatically detect those features associated with particular points of your mower, which is like your boundary of your lawn, so that next time it can be always contained within that boundary.

One of those things is not just following the boundaries or locating the location of your mower within the boundary. Furthermore, it can actually detect other features that can obstruct the movement or maybe affect the safety of this operation. For example, as you said, it can be used to avoid a collision with other obstacles or humans or any moving object.

It’s more flexible. It’s going to be safer than existing systems because this vision-based navigation system can not only provide the localization information of the mower … but also it’s going to provide much-needed safety features. It’s like a combination of a GPS navigation in your car … but it’s also providing a collision-avoidance system like modern cars.

DI: How long has this project been developed, and how did John Deere get involved?

SJC: Again, this is purely a technological development … Hopefully they will adapt our technology. Bottom line, they already have this product called the robotic mower introduced in Europe and other markets, but we are adding the system with a more advanced vision-based navigation system.

For that purpose, we were contacted by one of the engineers at John Deere because based on our background, … they said they could apply this same technology to the robotic mowers.

DI: When will this system be available on the market for the robotic lawn mowers?

SJC: I don’t know; hopefully soon, maybe in a few years. I shouldn’t know; I cannot speak on behalf of John Deere. This is really our research area, and we have a lot of technologies being developed from our group and other groups.

I believe this vision-based navigation system will really facilitate or speed up the adoption of this kind of robotic mower as mainstream household robots you can buy in addition to vacuum robots … in maybe five years, I hope.

[email protected]