Around the Corner or In the Home: Improving Robot Navigation and Localization

Imagine being dropped off in a new city without a map or your GPS.  Think about how you would move around and not get lost. Most people would try to recognize landmarks and keep track of their turns and motions, so they can eventually reach their destination or can retrace their steps. Now, imagine doing this with your eyes closed. That's what it feels like to be a robot without "eyesight."

Assitant Professor, Jana Kosecka

The problem of navigation and localization is one of the main challenges robots or autonomous cars must solve very accurately, and Dr. Jana Košecká's research seeks to open the "eyes" of robots allowing them to perceive the visual world. A person who is blind can learn and adapt because of the human brain, but it is even harder for a robot that is limited by its computer programming. The challenge for computer scientists is to build a sufficiently robust robot "brain" that can interpret what its cameras and sensors detect.

 "A really good example of computer vision at work is a robotic office or home assistant doing various fetch-and-delivery tasks such as navigating a room and trying to find keys, a phone, or a stapler," says Košecká. "In addition to knowing its way around, the robot assistant must be able to detect and recognize objects in large amount of clutter. Computer vision makes all this happen."

The use of visual information, however, goes beyond the navigation, recognition, or scene understanding tasks, which are the main problems that Košecká's group is focusing on. Developing computer programs for interpreting the visual world that surrounds us can benefit the visually impaired, enhance manufacturing and service robotics both in households and healthcare environments, expand robot use in search and rescue missions, or enable drones to deliver packages.

"In recent years our field made some great strides. Some pieces of the puzzle are coming together," says Košecká, "yet large portions remain unsolved. While we are still far away from having systems that can describe in words something as complex as Monet's paintings, the number of specialized applications that rely on interpreting visual information in restricted domains is growing faster then ever."

The natural environment is notoriously unpredictable and changing, so agents (robots) that can navigate around buildings or through landscapes in place of humans have valuable practical applications. Košecká says she finds the problems in her field contagiously interesting. She says, "I enjoy teaching students about Computer Vision, Robotics, Artificial Intelligence, and sharing a laboratory with many cool robots."

Learn more about Dr. Košecká's teaching and research.