“Choose a job you love, and you will never have to work a day in your life”, Confucio said, and Luca Carlone has certainly found a job he loves. He came to LIDS in July of 2015 as a post-doctoral associate. His work is focused on robot perception and on the particular challenges of Simultaneous Localization and Mapping (SLAM), which is—to put it simply—when a mobile robot builds a map of a new environment while concurrently using that map to move through it. Listening to Luca talk it’s clear that he never gets tired of his work: “The nice thing about robotics is that you can find so many different aspects that you never get bored.” If he needs a break from the mathematics involved in creating SLAM algorithms, he steps away to build a robot, write some code, or fly a small drone.
Originally from Italy, Luca was first introduced to robotics as an undergraduate at the Polytechnic University of Turin. Since then, he has been fascinated by industrial robotics such as the machines that assemble cars in a factory, an interest that led him to study mobile robotics—the self-driving car, for instance. He always wanted to make a positive impact on the world through his research, and this is why he focused on applied robotics. After getting his bachelor’s degree in Mechatronic Engineering in 2006, he received master’s degrees from both the Polytechnic University of Turin and the Polytechnic University of Milan. He then got his Ph.D. in Robotics from the Polytechnic University of Turin in 2012 before moving to the United States to explore the burgeoning opportunities here. He collaborated for two years with Frank Dellaert at Georgia Tech’s Borg Lab, studying robotic perception and computer vision, then he headed north to MIT to work with Sertac Karaman. “It’s a great mix because my supervisor is an expert in motion planning and I am an expert in perception, so we are trying to combine things,” Luca says. Luca focuses on getting the map of the environment correct, Sertac on helping the robot navigate that environment, so that together they are addressing a wide range of the issues that can arise in applied robotics.
Although Luca is working on multiple research projects at any given time, at the moment he has two main projects. One of them he doesn’t much discuss as it involves military applications; the other is related to the application of robotic technology in very tiny robots. “[I’m] interested in the challenges connected to scaling down these robots,” he says. These challenges are quite complex: “They [the tiny robots] cannot carry much load, so you cannot put a lot of computation on board, or a lot of sensors. So it’s harder to make the robot aware of what’s in its environment.” These are a few examples of the challenges of designing algorithms that allow a robot to understand and build a model of the world around it. The resource constraints are tricky to overcome and, in some cases, even to explain because humans perform so many of these world-building calculations without realizing it. “It’s surprisingly hard to tell people how they do these tasks because a human does this kind of job all the time without any trouble, so you can imagine that you’re using your eyes, you’re using touch to figure out what’s around you…and you mix all these sources of information into a coherent representation of the environment.” Robots can do these things only to a limited extent. Additionally, as Luca points out, “they have to be able to do them efficiently and with limited information because the robot can’t spend hours number crunching before performing its intended function.”
These days the functions and applications for robots may seem endless, and that’s part of the fun of Luca’s work. While his Ph.D. focused on a standard view of robotics including the four Ds (robots perform Dangerous, Dirty, Dull, and Dumb tasks), now he looks at different motivations. The current norm with robots is using them to move goods in a warehouse, to dispose of bombs, or to explore space; what’s needed going forward includes applications for intelligent transportation (there’s the self-driving car again), pollution monitoring in the form of situationally-aware drones, and precision agriculture. Given the projected population explosion over the next 40 years, Luca believes that the agricultural systems we have in place today can only keep up with demand through technology. Robots can be designed to monitor crop growth (this was part of his research at Georgia Tech) and to spray pesticides, among other duties, in a way that even huge teams of people could not.
However, Google’s self-driving car indicates some of the perils in transitioning these types of robots into industry and unveiling them to the public. The “real maturity of a technology,” Luca points out, “is often hard to assess, and users’ perceptions and expectations are often not aligned with what the robot can do.” “These robotic technologies are a good idea always in the long term, meaning that, when the technology is ready, we will really see a boost in performance between a human driver and a self-driving car,” he says. “The thing about the perception of the maturity of the technology is a big issue. [With Google’s self-driving car], everything seemed to be solved, and the technology seemed to be in good shape and ready for market, but if you see the news from [late March], the same people from Google are saying that you have to be careful because in a very specific environment, the technology is close to being delivered, but there are still many open problems, and for some places around the globe, it may take 30 years.”
Key to all of Luca’s work, both now and going forward, is making the existing algorithms for robots more robust. “An algorithm is software that is processing a bunch of information and giving an answer. People get very excited when using algorithms, but…showing that your algorithm works in a single instance does not mean that it works in general.” Humans may be good at selectively filtering huge amounts of information and processing it appropriately with little conscious thought, but for robots, “the amount of information that you need to put in your map representation depends on the task that the robot has to do.” The goal, in other words, is not to have a robotics expert sitting in a room designing every possible application; it’s to create a robot smart enough to figure this out for itself. For this to happen, though, the expert must first be able to tell the robot what to watch for since it will have limits on what it can pay attention to, no matter how many sensors it has and how much data it can take in.
For Luca, LIDS is the perfect place to explore these questions. “Here,” he says, “the only real constraint is time.” There are so many great seminars and talks to attend, so many colleagues to pair up with for projects and for hanging out, that time management becomes a necessary skill. Luca also appreciates how easy it is to interact with people in other departments, partly because of how well-located LIDS is, but most of all he fully agrees with the mission of LIDS,: “to produce strong theoretical contributions which have a huge impact on real life. That’s exactly what I want to do: something that is grounded in scientific research but that becomes a real thing that helps people.”