Biomedical engineer René Vidal specializes in a field that blends biomedical imaging, computer vision and machine learning. As director of the Johns Hopkins Vision Dynamics and Learning Lab, he develops programs that give computers and robots an ability akin to human vision.
The human visual system is the result of millions of years of evolution. How can you produce a computer that can do the same? In basic technical terms, how do you go about this?
VIDAL: Segmentation is key. You start with an image of something. Then you write a program that separates, or segments, the image into different regions based on characteristics, such as intensity, color or texture. A lot of my research involves segmenting, finding the boundaries in an image of a tumor, a face, whatever. A surgeon might use a program like this to find the boundaries of a tumor in an MRI scan.
But this is not an easy problem. Algorithms can make mistakes.
Then, machine learning comes in. Segmentation algorithms can make mistakes. So we’re trying to refine the algorithms through machine learning. Basically, this means the user will interact with the algorithm. If a program shows the boundaries of an object that don’t seem quite right, the user can correct those boundaries. Through iterations of that process, the program provides better and better segmentations.
How did you choose this field of research?
VIDAL: I studied electrical engineering as an undergrad at the Catholic University of Chile, and I was very much interested in robotics. During my last year, I remember that a team from Carnegie Mellon went to Chile to test a robot that was going to be taken to the moon in the Atacama Desert. The team included a former student from the Catholic University, who came to give a talk about the project. Among many other sensors, the robot had a pair of stereo cameras that provided the robot with depth perception for navigation purposes. Greatly inspired by the talk and the project, I became very interested in the use of visual perception to navigate robots.
For my Ph.D. I wanted to control robots with visual sensors. My project involved getting unmanned helicopters to land automatically on moving platforms. The helicopters had cameras that could detect the platforms. They would fly over, identify their target and go down to land. We used model helicopters about 1 1/2 meters long. But unmanned vehicles are being used for a variety of purposes, such as traffic monitoring and search and rescue operations.
After I came to Johns Hopkins, I became interested in the biomedical applications of visual image analysis.
Do you envision a future where robots will be used for everything? What is the benefit to making robotics more human when humans have such unique skills?
VIDAL: I envision a future where robots will be used for everything, especially to help humans. For instance, we have already seen the vacuum cleaner robot in our homes, the da Vinci robot in the operating room, underwater robots that can find the black box of an airplane in the middle of the ocean. In the future, such robots will be able to perform such tasks much more precisely and in collaboration with humans. For example, we will see robots that can cook, help the elderly at home, and so on.
--Interviewed by Melissa Hendricks