For Georgia to produce its over 130 million pounds of peaches ever year, farmers and workers around the orchard are tasked with a set of important responsibilities: pruning, thinning, and picking.

Researchers at the Georgia Tech Research Institute (GTRI) are testing out a new idea on the University of Georgia’s Griffin campus.: Have a robot do all three.

An intelligent robot being developed at The Georgia Tech Research Institute (GTRI) uses artificial intelligence and sophisticated navigation to handle the tasks of removing branches, finding good peaches, and removing them from a tree. The automated, non-human approach could result in significant cost savings for peach farms in Georgia, according to the researchers.

"This is something that directly affects the yield of the trees," said Gary McMurray, a GTRI principal research engineer and division chief of GTRI's Intelligent Sustainable Technologies Division in a recent press release . "It's money in the pocket of the growers."

The Georgia Tech-developed robot uses both LIDAR and GPS to self-navigate through the orchard. The LIDAR system determines distances by targeting an object with a laser and measuring the laser beam's return time, while the GPS technology measures locations as specific as a fraction of an inch.

The engineers' efforts are focused mainly on the tasks of thinning and pruning, rather than harvesting.

Researchers trained the system to recognize undeveloped peaches known as peachlets. Removing the peachlets, a process often referred to as thinning, is critical to the output of a given tree.

Thinning allows for bigger and better peaches to grow, according to Ai-Ping Hu, a GTRI senior research engineer who is leading the robot design project.

"If you just let all the peaches grow to maturity, then what you'll end up getting is a tree of really small peaches," Hu said. "What you want to do is have relatively few peaches, but you want the ones that remain to be nice and big and sweet – ones you can actually sell."

Once at a peach tree, the robot's embedded 3D camera finds the peachlet and grabs the fruit with a claw-like device at the end of its robotic arm.

The robot is also trained to assist with pruning, the selective removal of branches prior to the spring growing season, which typically occurs from mid-May to early August. The important process, usually done by hand, exposes more interior surface areas of the fruit trees to sunlight.

None of today's robots have been able to fully replace humans in the peach cultivation industry due to peach orchards' unstructured environments, which includes unpredictable weather, uneven terrain, and trees' different shapes and sizes, said Hu.

The robot utilizes a combination of LIDAR remote sensing and GPS technology to self-navigate through a peach orchard. Once the robot arrives at a peach tree, it uses an embedded 3D camera to determine which peaches need to be pruned or thinned. (Credit: Ai-Ping Hu, Georgia Tech Research Institute)

"There's no robot in the world right now that can harvest or thin peaches as well as people can," Hu said. "The technology's not quite there yet."

In a short Q&A with Tech Briefs below, Hu explains why it's so important to improve robotic capabilities so they can handle such unstructured environments.

Tech Briefs: What have been the shortcomings of earlier methods of automating the harvesting process with robots?

Ai-Ping Hu: The focus, and novelty, of the project is primarily peachlet thinning, rather than harvesting. Compared to fruit harvesting, there has been very little prior work devoted to thinning (which has roughly the same manual labor requirements). From a robotics standpoint, it is a more difficult problem because the peachlets are smaller and their coloration is similar to their surroundings.

Tech Briefs: What characteristics does your 3D camera recognize in order to determine a peach that’s ready to be removed?

Ai-Ping Hu: We trained convolutional neural networks (deep learning) to recognize peachlets. The RGB-D (color plus depth) camera can provide 3D information to the robot, which then uses a technique called visual servoing to close the feedback loop that permits co-locating the end-effector claws to the peachlet.

Tech Briefs: Can you walk me through an example of how the robot operates in a sample application of finding and pulling and pruning a peach, from start to finish?

Ai-Ping Hu: Each tree in the orchard has a known location (via RTK-GPS , which is much more accurate than regular GPS). The robot platform autonomously navigates to a given tree, avoiding obstacles along the way. The robot arm is then deployed to perform peachlet thinning.

A good video can be found here below:

Tech Briefs: How have farmers reacted to these demonstrations? Is the idea to fully replace humans in this task, or to have the robot work alongside humans?

Ai-Ping Hu: Our mobile robot arm system is a research platform. It has been developed in consultation with horticultural researchers at the University of Georgia Griffin campus. The intent is that the eventual robot platform will operate autonomously. The selection of the off-the-shelf Universal Robotics UR5 collaborative robot  enables it to operate safely around people.

Tech Briefs: How does the time and production of a robot compare to a human worker?

Ai-Ping Hu: No existing robot can match human speeds yet. But some are getting close for harvesting and there are small agricultural robotics startups newly out there.

Tech Briefs: How does the robot handle tough terrain? And high branches?

Ai-Ping Hu: The mobile base has large-tread inflatable tires to easily roll over roots and debris. For our testing, we have been focused on low-hanging peachlets.

Tech Briefs: And what’s next, regarding this work?

Ai-Ping Hu: We are currently exploring pruning, which is the selective removal of branches prior to the growing season. We are both trying to model the process and provide automated suggestions for the farmer based on in-situ solar information and also automating the actually cutting with a new end-effector on the same mobile robot arm platform.

What do you think? Share your questions and comments below.