Give us the overview of GROWBOT and what you plan to accomplish through your research.
GROWBOT is the much catchier acronym for my project “A Grower-Reprogrammable Robot for Ornamental Plant Production Tasks”. The top-level mission for the project is to investigate ways in which non-expert users (i.e., those without technical expertise in robot programming and control), but that are nevertheless skilled in plant processing, can use robots in their work, to relieve them of the more repetitive, labor-intensive tasks encountered.
I am doing this research towards a PhD in Robotics in the Robot Learning Lab at King’s College London, a research group led by Dr. Matthew Howard that specializes in machine learning for control, especially as applied to robotics.
My project is sponsored by a group called the Agriculture and Horticulture Development Board (AHDB), who represent the research and development interests of farmers and growers in the UK. They are strong supporters of this project, given an increasing issue in securing seasonal labor in the UK; though that said this problem is not exclusive to the UK, with similar problems seen in the Netherlands, Australia and the U.S.
What excites you most about this project?
There are many aspects to the project that have made it exciting to work on – for starters, there’s the great puzzled look when I tell people I work on flower-picking robots! I think really what is a great motivator has been meeting with growers working in the industry and seeing the challenges they are facing first-hand. I’ve spoken with growers in the UK, Netherlands, the US, and Australia to help gain an overall understanding of how the industry works, and get a grasp on the very real growing problems they are now facing in securing labor to do the huge variety of manual tasks that are required.
From a robotics point of view, the tasks which growers have struggled to automate actually provide really interesting research challenges. Handling plants involves a great amount of uncertainty in sensing, and there’s a reasonably high performance requirement in terms of through-put. Additionally, growers typically will have several varieties in production to meet consumer demands in relatively small batch sizes, and will have different requirements for packaging and presentation for each consumer (a challenge which can come up even in high-volume production), adding further challenges to automation.
Given these challenges, ultimately growers need flexible automation that can adapt to the challenges they are facing that month/week/day. To this end, another thing which I am excited about is that this project has let me explore the field of Programming by Demonstration (aka Learning from Demonstration/Imitation Learning/etc. depending on who you’re talking to and what you’re doing), a field of research that has long captured my imagination as a brilliant intersection between machine learning, control, and people.
I remember during my Undergraduate degree seeing videos like Petar Kormushev’s classic pancake flipping robot and being blown away by the brilliant “simplicity” of it – get the robot to do all the hard work through machine learning! I’ve since realized it’s all far from simple (making it all the more interesting), and I’m very happy to have found myself doing a PhD on the topic with a supervisor who has a lot of experience in the area (Dr. Matthew Howard).
A recent exciting outcome from my work in this area was publishing my first paper, “Teaching Human Teachers to Teach Robot Learners”, which I presented in Brisbane this May at the International Conference on Robotics and Automation. This paper is a first step towards my project goal of letting non-expert users deploy robots for increasingly complex tasks, and I’m currently working on a body of work that extends this paper to more real-world tasks where Sawyer was tested directly working with various plants.
How does a Sawyer cobot fit in?
As you might be able to guess, the need for a flexible automation solution which can be adapted to the paradigm of programming by demonstration, and is generally safe to be in close proximity to people, means that a Sawyer cobot is a very good fit for this project. I’ve been using the SDK extensively for this project, and thanks to the hard work of Ian McMahon and Co. it’s been a relatively stress-free experience from the start. Having worked with a variety of robots in the past, it was a relief having ROS baked in!
How did you become interested in the fields of robotics and engineering? Was it something your parents introduced you to or did you discover it through school when you were younger?
My parents certainly helped influence me down the path toward STEM in general, back in the 90’s my dad worked for Emirates as an aircraft engineer meaning I got to occasionally get a backstage tour of stripped out airplanes and poke my head in wings and engines. I’ve got early memories from school when I was about 5 where we had to program one of those grid-robots (move forward, turn left, move forward, etc.), and later in my teens I participated in a FIRST Lego robotics rescue robot competition. All of this helped set a direction, but I would say I went from having a general interest in robotics to a more fully-fledged obsession during my undergraduate in mechanical engineering at Trinity College Dublin, Ireland, when I met (then) PhD student Conor McGinn and my project supervisor Kevin Kelly.
Robotics at the time was not a well-developed field in Ireland, so it was a bit of trial-by-fire building an obstacle-avoidance system for a Kinect guided robot for my thesis! It was early days for the Kinect – I still get nightmares of trying to get everything to compile! Fortunately, it worked out well and after a little time in industry with National Instruments I rejoined the group for a Masters where I worked on a human robot interaction project for assistive service robots. The robotics group in TCD has since grown from a small group of initially about 5 people when I first joined in 2010 to about 15-20 people between PhD, Masters, and Undergraduate students where they’ve been doing great work in assistive service robotics.
During my time in King’s, I’ve also done a little bit of “paying it back” by running school robotics workshops using Lego Mindstorms robots, with projects angled around agriculture and horticulture.
What do you see yourself doing five years from now?
This is the big, tricky question. There is a lot of industry interest in my work, so it would be great to see it evolve to real-world use, and I’ve got a bunch of research ideas of my own that I’d like to work on, but my future planning capacities have been somewhat absorbed by the demands of the PhD – I think I have the next year just about figured out, five years is future-Aran’s problem for the time being! I can say that while the future is uncertain, the capabilities of AI and robotics are improving at a fantastic rate and I’m hoping I can continue leveraging them to address the big challenges of the day.
Next, take a deeper dive into Aran’s work at aransena.com and if you’re the Twittering type, connect with him at @aransena. Or if you prefer to stick around here, check out the rest of the Rethink Robotics blog and sign up above to get alerts sent directly to your inbox.