Imagine a robot executing an order such as finding a coral head in an ocean. This is one of the projects Dr. Florian Shkurti, an assistant professor of computer science at UTM, is working on. Shkurti, who researches “mobile robotics, computer vision, [and] machine learning, planning and control,” also serves as a faculty member at the University of Toronto Robotics Institute and a faculty affiliate at the Vector Institute alongside teaching at UTM. In this issue, Shkurti discusses how he was inspired to pursue this field, his research on self-driving and field robotics, and the future of robotics.
Shkurti’s passion for computer science stems from an introductory high school programming class where he discovered his interest in logic. As an undergraduate computer science and mathematics student at the University of Toronto, Shkurti originally aspired to become a video game designer. However, his plans soon changed, when in his third year at U of T, he was introduced to Rafaello D’Andrea, a Canadian/Italian/Swiss engineer, artist, and entrepreneur.
D’Andrea’s unique integration of robotics and art was highly inspiring for Shkurti. “I wanted to become as technically proficient and as fearless in my choice of projects as [D’Andrea],” Shkurti remarks. With this newfound passion, Shkurti aimed to “program machines that moved and reacted to what they saw.” Furthermore, despite the limited resources and opportunities available in the computer science field at the time, Shkurti continued to pursue the field, consistently excelling as demonstrated by his numerous awards and publications.
Equipped with skills gained by working in the Mobile Robotics Lab at McGill University, Shkurti is currently working on enabling “robots to learn from their own experience and from demonstrations.” His project examines reinforcement learning which is essentially “learning by trial and error.” It also delves into imitation learning in which one learns from another’s observations and experiences in learning.
“I think [imitation learning] is the key to making robots interact efficiently with humans who might not have the technical background to actually program machines,” explained Shkurti. “By guiding or affecting the behavior of their robots by a small number of demonstrations and maybe corrections, [people may] not have to write code. Robots should be able to understand the objectives of their users with as little interaction as possible.”
Another project of Shkurti’s involves self-driving car simulators. Typically, simulations are run by humans who test various scenarios. However, Shkurti’s team is researching “ways to automatically search the space of possible adversarial behaviors of pedestrians, other cars, buildings, and scene appearances that will cause the software of a self-driving car to fail.”
Shkurti emphasizes the practicality and wide-spread application of robotics. His research includes field robotics, which entails operating robots in unstructured and natural environments such as forests, oceans, and rivers. This operation requires teaching robots how to incorporate human feedback to better visually explore their environment.
“I want to put robots in the service of biologists and environmental scientists who need to visually search expansive environments for particular features,” said Shkurti. Ideally, these robots would be able to take an order such as: “Find me a live coral head in the ocean” and search the region to identify live coral heads. Shkurti compares it to how a Google image search could process that order to display an image of a live coral head.
Mitigating problems such as visual navigation and visual attention are factors being investigated currently. Shkurti feels the urgency of research in this particular field of robotics, especially because biologists could significantly benefit with the support of robots. However, challenges, like knowing how to create systems biologists can easily and efficiently use, still exist.
Although the future and need for robotics is a contentious issue, Shkurti predicts that robot use will increase, especially in manufacturing. Moreover, he hopes there are increased avenues for medical robotics as the potential benefits of such applications are numerous, particularly within large hospital settings. Conversely, fields such as self-driving cars likely need more time to reach the optimal level of reliability. “There’s a lot of good that can come out of robotics and computer science if people focus on the right p