This article has been corrected. You may view this article’s correction here.

AK Watson can tell if your tone of voice is angry or happy, but not what your words mean.

Possessing a “brain” made up of 16 computers connected over a network, Watson is the latest humanoid robot under development in computer science professor Brian Scassellati’s lab.

The robot is designed to evaluate models of social development in children and help diagnose disorders like autism. Scassellati’s creation is just one of the many projects researchers in the Computer Science Department intend to develop into practical applications. Others include the creation of realistic computer graphics and a bug-free software development system.

AK Watson has a head, neck, torso, one arm and 20 joints and motors. It is the size of a one-year-old child, mimics the movement of an infant, and is modelled after theories about how children learn to reach and grasp objects, Scassellati said.

“It has basic hand-eye coordination and can recognize other individuals as distinct from itself,” he said. “It can recognize itself in a mirror.”

Watson possesses simple perceptual systems and can also identify the location and activities of nearby people, as well as the emotion intonated in their voices, Scassellati said.

He said his lab, which includes five graduate students and four undergrads, is working with the Yale Child Study Center at the School of Medicine to use robots in developing more quantitative clinical diagnoses for autism in children.

“Right now, autism is diagnosed by having trained clinicians observe and play with children, which can be very subjective,” Scassellati said. “People generally don’t have a lot of quantitative data for evaluating autism, and we’re trying to develop technology in the form of robots and smart rooms with hidden cameras and microphones so that we can get that data.”

Scassellati said he has been working on humanoid robot projects at Yale for almost four years and does not have a specific endpoint in mind.

“We’re never really going to be done, since we are trying to learn about kids,” he said. “There will be different versions of hardware coming out, but we won’t really be finished until we know all there is to know about social development.”

He said the lab has consulted with commercial developers but does not intend for the robots to be marketed commercially.

In contrast, Zhong Shao, director of undergraduate studies for the Computer Science Department, said the purpose of his research is to fundamentally change commercially marketed software.

The FLINT project, under development since 1997, aims to develop technology to ensure that all software will be reliable and free of bugs and security holes, Shao wrote in an e-mail. When consumers download or purchase software nowadays, he said, there is no real guarantee from developers that the software is secure.

“The FLINT project aims to do something completely new,” he said. “Instead of fixing buggy software after the fact — which is what the industry is doing today — we develop new technologies that allow people to write what we call ‘certified software.'”

Certified software would include a rigorous logical proof that could be checked by a small “proof-checker” program, along with the usual executable program, Shao said.

“Many people would think that this is a crazy idea at the first blush, since how many of us have actually written a rigorous proof when writing a program?” he said. “They would think that this would never be practical. [But] it turns out that every programmer, if she is interested in making her code reliable, is implicitly already doing such reasoning.”

Shao said the FLINT project will help programmers convert their informal and mental proofs into machine-checkable ones. The proofing system would eliminate most errors, but hardware defects could still be a problem, he said.

Though the project is still in its early stages, Shao said, his lab’s long-term goals include commercializing the new technology.

“There is already huge demand for high-confidence software,” he said. “Once we can show that such technology exists and is practical, I believe that the industry will embrace it.”

More recent than the humanoid robot and FLINT projects, research conducted by computer science professors Holly Rushmeier and Julie Dorsey over the past year has dealt with developing new applications for computer graphics.

Rushmeier said their graphics lab undertakes a number of interrelated projects.

“Our work is motivated by practical problems,” she said. “We ask, ‘how can people make graphics more efficiently?’ and our projects range from traditional applications, as in the graphics used in feature films and computer games, to architectural design and cultural heritage documentation.”

She said the researchers are finding new ways to create and edit geometric models from two-dimensional sketches. They are also working on projects aimed at faithfully reconstructing the appearance of a historical places and artifacts.

The goal of the lab’s projects, Rushmeier said, is not to bring new products to market, but to do basic research into technology that will eventually have applications, commercial and otherwise.

“Our overall goal is to make graphics easier for people to use and create,” she said. “We would hope our developments could eventually become widely available.”

Rushmeier said she and Dorsey do not develop computing hardware, but they assemble their own graphics systems and experiment by using scanners and lights in different combinations.

Other ongoing research in the Computer Science Department involves artificial intelligence, machine learning, mathematical theories of human vision, cryptography and programs for physical- or biological-science research.
[ydn-legacy-photo-inline id=”16002″ ]