Precision Agriculture: Sustainable Farming in the Age of Robotics
Early in the fall of 2008, students began gathering before a raised platform of fake grass. The artificial turf was adorned with evenly spaced tomato plants, nestled in sensible terra cotta pots. And while the small cluster of plants and grow lamps might have seemed incongruous under other circumstances, this was a garden with a twist: instead of being horticulturalists, the humans were there only to program and supervise. The caretakers of the plants would be entirely robotic.
Luke Johnson and Sam Dyar program an autonomous robotic arm
Photo: Jason Dorfman
Photo: Jason Dorfman
The idea for the project came from work done by Nikolaus Correll, a postdoctoral assistant working in Professor Daniela Rus’ Distributed Robotics Lab. Correll, who came to CSAIL in 2007, saw the possible applications of swarm robotics to an agricultural environment. In the long view, the researchers hope to develop a fully autonomous greenhouse, complete with robots, pots and plants connected via computation, sensing and communication. Each robot is outfitted with a robotic arm and a watering pump, while the plants themselves are equipped with local soil sensing, networking, and computation. This affords them the ability to communicate: plants can request water or nutrients and keep track of their conditions, including fruit produced; robots are able to minister to their charges, locate and pick a specific tomato, and even pollinate the plants.
The system, which Rus refers to as precision agriculture, has a double advantage over the way crops are currently cultivated and harvested. First, due to each plant’s ability to monitor and broadcast its own physical state, water, nutrients and care will be dispensed on an as-needed basis. This kind of targeted specificity should allow for a great reduction in resources consumed in the growth process, ameliorating the heavy carbon footprint of today’s agriculture. Furthermore, a mechanical harvest removes the backbreaking work currently involved in reaping specialty crops such as fruit and vegetables. (By contrast, broad land crops such as small grains and hay are already being harvested in a way that is at least partially mechanized.) The DRL is no stranger to work that fuses the robotic with the biological; other projects include mobile networks of underwater robots and cow herding with virtual fences. One quirk of this project is that while part of the research was undertaken during IAP in January, much of the work was begun by undergraduates organized into specialized teams to build a distributed robotic garden within the time frame of a course. This specificity required them to find innovative ways to develop their ideas and work together across project areas. The course structure that Rus and Correll developed allowed for the tasks (such as object recognition or navigation) to be addressed in great detail by a handful of students. What the pair found, as the class went on, was that the small, specialized teams were serving a dual purpose. Not only did the specific groups allow for greater concentration on discrete tasks, but communication between groups led to a surprisingly high level of information sharing. The result? The combination of specialization and communication led to a nuanced collective understanding of the project as a whole that might have been difficult to obtain in any other way. This collaborative approach to robotics is part of a larger shift going on in the field. In the past five to ten years, the field has seen increasing cooperation on projects of greater and greater complexity. Frequently, the tools generated in this process can be recycled for use in still more work, creating a kind of open source community of roboticists. In the garden project, for example, the students were able to use several projects already under development in CSAIL. A tool called LCM, or Lightweight Communications Marshaller, was used to allow the different robotic modules to communicate; the version used in the project came from the DARPA Grand Challenge Vehicle. The object recognition is built on the back of LabelMe, an image annotation tool pioneered by Bryan Russell and Professors Antonio Torralba and Bill Freeman. And the robotic bases themselves? Re-imagined versions of iRobot’s Roomba. Ultimately, the project’s goal is to foster excitement in the roboticists of the future about what is possible; in this, it has succeeded with flying colors. At its end, there was an incredibly high level of enthusiasm about the things accomplished during the semester and the work ahead. Students from nearly every specialized group chose to stay on for the January interterm period to develop their findings further. Correll is optimistic about future applications of the project and others like it. Looking past agriculture, he ruminated on other tasks to which this sort of system can be applied down the line. Tasks, for example, like automating services for older adults with low mobility in residential care facilities, or tending to plants in greenhouses or hydroponic farms. The technology, once perfected, is immensely adaptable, and hints at an exciting future of collaboration between humans, the natural world, and our machines.
April 22 2013
Adwoa Gyimah-Brempong, CSAIL
Adwoa Gyimah-Brempong, CSAIL