Georgia Tech Team Wins New Fetch Robot at ICRA's FetchIt! Mobile Manipulation Challenge
Jun 28, 2019 — Atlanta, GA
Sonia Chernova’s Robot Autonomy and Interactive Learning (RAIL) lab is adding a new member this summer after a successful foray into the FetchIt! Mobile Manipulation Challenge at the International Conference on Robotics and Automation (ICRA) last month.
A team of Georgia Tech master’s and Ph.D. students, advised by Chernova, won the challenge by successfully assembling three kits with its robot in 39 minutes. It was the only team in the competition to complete the task, with the second-place finisher failing to score a point.
For its victory, the RAIL lab will receive a new mobile manipulation robot from Fetch Robotics, its second. Along with the other robots already in the lab’s possession, the newcomer will provide RAIL researchers new opportunities to pursue multi-robot applications. The prize package also includes items from the event’s co-sponsors EandM Robotics, Schunk, SICK Sensor Intelligence, and The Construct, to go with the $100,000 robot.
“This is a long-term benefit,” said Chernova, an associate professor in the School of Interactive Computing. “This is one of the most capable mobile manipulation platforms out there, and to now have two of them will enable us to enhance the capabilities of the robot and pursue new lines of research in our lab.”
The allure of a new state-of-the-art robot would be enough to entice most teams to take part in the competition, but for Chernova and her participating students it was more about the opportunity to explore specific applications that aligned with their research initiatives, past and present.
The lab has done past work in grasping, semantic reasoning and mapping, and fault diagnosis, the latter of which has become a focus over the past six months. The competition, Ph.D. student David Kent said, came at a good time because the particular challenges it presented are often in this domain.
“This particular setup was particularly challenging because there was just enough variability where it wasn’t going to work every time,” he said. “There would always be something going wrong, so fault recovery ended up being very central.”
To win the competition, not only did Georgia Tech’s team have to come in first place, it had to do so by scoring at least 14 points. To put that into context, Georgia Tech was the only team in the competition to finish with any points. Teams scored points by successfully collecting items laid out at different stations to assemble three kits. They were awarded eight points for each completed kit. Any kit that was missing a piece, however, resulted in zero points awarded, and any kit with extra pieces would have points deducted.
“If you drop one screw along the way and you don’t notice – which is actually very easy to do – you go away with nothing,” Chernova said. “In the real world, a partial kit is useless.”
Georgia Tech achieved its first 15 points and elected to complete its third kit without official scoring to ensure it wouldn’t drop below the threshold needed to win the robot. Officially the team scored 15, but a completed third kit gave it an unofficial 23 points, after bonuses were added.
“It was a lot of fun to be able to work with my lab on a single project and see it come together,” said Ph.D. student Weiyu Liu, another member of the team. “It was a really great opportunity to try out some of the code we had written and also to see others’ code and other research projects.”
Already, the team has turned the experience into a submitted paper, which they hope to have accepted and published in the future. The focus is on mobile manipulation, which is a particularly challenging aspect of robotics because of what Chernova calls “an explosion of uncertainty.”
“Manipulation in many ways is a solved problem,” she said. “Navigation in many ways is a solved problem. When you put those two solved problems together, though – when you take the wheels and put the arm on it – it becomes a much more challenging problem, one our research will continue to tackle with the aid of Fetch in the coming years.”
Members of the team included: Chernova, Kent, Liu, Siddhartha Banerjee, Angel Daruna, Jonathan Balloch, Abhinav Jain, Akshay Krishnan, Muhammad Asif Rana, Harish Ravichandar, Binit Shah, and Nithin Shrivatsav.