Developing Robots That Can Teach Humans

Engineerblogger
March 6, 2012

A few years ago, AnthroTronix, Inc., an engineering research and development firm in College Park, Md., introduced Cosmobot, a type of social robot for therapists and educators who work with developmentally and learning disabled children, including those with autism and cerebral palsy. By imitating human joint movement in its shoulders, arms, hands and head, Cosmobot motivates children to develop new skills more quickly than is typical with traditional therapy. But why does this work? Why do children respond so favorably to educational programs taught by technology? And, when the technology is a robot made from inanimate materials, how do children learn to distinguish between the robot and a living thing? Credit: NSF

Researchers are programming robot teachers to gaze and gesture like humans

When it comes to communication, sometimes it's our body language that says the most--especially when it comes to our eyes.

"It turns out that gaze tells us all sorts of things about attention, about mental states, about roles in conversations," says Bilge Mutlu, a computer scientist at the University of Wisconsin-Madison.

Mutlu knows a thing or two about the psychology of body language. He bills himself as a human-computer interaction specialist. Support from the National Science Foundation (NSF) is helping Mutlu and his fellow computer scientist, Michael Gleicher, take gaze behavior in humans and create algorithms to reproduce it in robots and animated characters.

"These are behaviors that can be modeled and then designed into robots so that they (the behaviors) can be used on demand by a robot whenever it needs to refer to something and make sure that people understand what it's referring to," explains Mutlu.

Both Mutlu and Gleicher are betting that there will be significant benefits to making robots and animated characters "look" more like humans. "We can build animated agents and robots that can communicate more effectively by using the very subtle cues that people use," says Gleicher.

Mutlu sets up experiments to study the effect of a robot gaze on humans. "We are interested in seeing how referential gaze cues might facilitate collaborative work such that if a robot is giving instructions to people about a task that needs to be completed, how does that gaze facilitate that instruction task and people's understanding of the instruction and the execution of that task," says Mutlu.

To demonstrate, a three-foot-tall, yellow robot in the computer sciences lab greets subjects, saying: "Hi, I'm Wakamaru, nice to meet you. I have a task for you to categorize these objects on the table into boxes."

In one case, the robot very naturally glances toward the objects it "wants" sorted as it speaks. In another case, the robot just stares at the person. Mutlu says the results are pretty clear. "When the robot uses humanlike gaze cues, people are much faster in locating the objects that they have to move."

Another experiment run by Mutlu and Gleicher's team explores how an animated character's eyes affect human learning. A character projected on a screen says to the viewer, "Today, I'll be telling you a story that comes straight from ancient China." Behind the animated character is a map of China that he'll be referring to in the lecture that runs several minutes.

"The goal of the experiment is to see if we could achieve a high-level outcome, like learning, by controlling an animated character's gaze," says Gleicher. "What we found was when the lecturer looked at the map at appropriate times to indicate to the participant that now I'm talking about something on the map, the participant ended up learning more about spatial locations."

The team hopes their work will transform how humanoid robots and animated characters interface with people, especially in classrooms. "We can design technology that really benefits people in learning, in health and in well-being, and in collaborative work," notes Mutlu.

Now, that's technology worth keeping an eye on!




Source:  National Science Foundation (NSF)

Discovery of a ‘Dark State’ Could Mean a Brighter Future for Solar Energy

Engineerblogger
Dec 16, 2011

Professor Xiaoyang Zhu

The efficiency of conventional solar cells could be significantly increased, according to new research on the mechanisms of solar energy conversion led by chemist Xiaoyang Zhu at The University of Texas at Austin.

Zhu and his team have discovered that it's possible to double the number of electrons harvested from one photon of sunlight using an organic plastic semiconductor material.

"Plastic semiconductor solar cell production has great advantages, one of which is low cost," said Zhu, a professor of chemistry. "Combined with the vast capabilities for molecular design and synthesis, our discovery opens the door to an exciting new approach for solar energy conversion, leading to much higher efficiencies."

Zhu and his team published their groundbreaking discovery Dec. 16 in Science.

The maximum theoretical efficiency of the silicon solar cell in use today is approximately 31 percent, because much of the sun's energy hitting the cell is too high to be turned into usable electricity. That energy, in the form of "hot electrons," is instead lost as heat. Capturing hot electrons could potentially increase the efficiency of solar-to-electric power conversion to as high as 66 percent.

Zhu and his team previously demonstrated that those hot electrons could be captured using semiconductor nanocrystals. They published that research in Science in 2010, but Zhu says the actual implementation of a viable technology based on that research is very challenging.

"For one thing," said Zhu, "that 66 percent efficiency can only be achieved when highly focused sunlight is used, not just the raw sunlight that typically hits a solar panel. This creates problems when considering engineering a new material or device."

To circumvent that problem, Zhu and his team have found an alternative. They discovered that a photon produces a dark quantum "shadow state" from which two electrons can then be efficiently captured to generate more energy in the semiconductor pentacene.

Zhu said that exploiting that mechanism could increase solar cell efficiency to 44 percent without the need for focusing a solar beam, which would encourage more widespread use of solar technology.

The research team was spearheaded by Wai-lun Chan, a postdoctoral fellow in Zhu’s group, with the help of postdoctoral fellows Manuel Ligges, Askat Jailaubekov, Loren Kaake and Luis Miaja-Avila. The research was supported by the National Science Foundation and the Department of Energy.


Source: University of Texas at Austin

Additional Information:

• In study "Observing the Multiexciton State in Singlet Fission and Ensuing Ultrafast Multielectron Transfer" by Wai-Lun Chan, Manuel Ligges, Askat Jailaubekov, Loren Kaake, Luis Miaja-Avila, and X.-Y. Zhu in Science.

Vision Scientists Demonstrate Innovative Learning Method

Engineerblogger
Dec 12, 2011

In the future, a person may be able to watch a computer screen and have his or her brain patterns modified to improve physical or mental performance. Researchers say an innovative learning method that uses decoded functional magnetic resonance imaging could modify brain activities to help people recuperate from an accident or disease, learn a new language or even fly a plane.  Credit: Nicolle Rager Fuller, National Science Foundation

New research published today in the journal Science suggests it may be possible to use brain technology to learn to play a piano, reduce mental stress or hit a curve ball with little or no conscious effort. It's the kind of thing seen in Hollywood's "Matrix" franchise.

Experiments conducted at Boston University (BU) and ATR Computational Neuroscience Laboratories in Kyoto, Japan, recently demonstrated that through a person's visual cortex, researchers could use decoded functional magnetic resonance imaging (fMRI) to induce brain activity patterns to match a previously known target state and thereby improve performance on visual tasks.

Think of a person watching a computer screen and having his or her brain patterns modified to match those of a high-performing athlete or modified to recuperate from an accident or disease. Though preliminary, researchers say such possibilities may exist in the future.

"Adult early visual areas are sufficiently plastic to cause visual perceptual learning," said lead author and BU neuroscientist Takeo Watanabe of the part of the brain analyzed in the study.

Neuroscientists have found that pictures gradually build up inside a person's brain, appearing first as lines, edges, shapes, colors and motion in early visual areas. The brain then fills in greater detail to make a red ball appear as a red ball, for example.

Researchers studied the early visual areas for their ability to cause improvements in visual performance and learning.

"Some previous research confirmed a correlation between improving visual performance and changes in early visual areas, while other researchers found correlations in higher visual and decision areas," said Watanabe, director of BU's Visual Science Laboratory. "However, none of these studies directly addressed the question of whether early visual areas are sufficiently plastic to cause visual perceptual learning." Until now.

Boston University post-doctoral fellow Kazuhisa Shibata designed and implemented a method using decoded fMRI neurofeedback to induce a particular activation pattern in targeted early visual areas that corresponded to a pattern evoked by a specific visual feature in a brain region of interest. The researchers then tested whether repetitions of the activation pattern caused visual performance improvement on that visual feature.

The result, say researchers, is a novel learning approach sufficient to cause long-lasting improvement in tasks that require visual performance.

What's more, the approached worked even when test subjects were not aware of what they were learning.

"The most surprising thing in this study is that mere inductions of neural activation patterns corresponding to a specific visual feature led to visual performance improvement on the visual feature, without presenting the feature or subjects' awareness of what was to be learned," said Watanabe, who developed the idea for the research project along with Mitsuo Kawato, director of ATR lab and Yuka Sasaki, an assistant in neuroscience at Massachusetts General Hospital.

"We found that subjects were not aware of what was to be learned while behavioral data obtained before and after the neurofeedback training showed that subjects' visual performance improved specifically for the target orientation, which was used in the neurofeedback training," he said.

The finding brings up an inevitable question. Is hypnosis or a type of automated learning a potential outcome of the research?

"In theory, hypnosis or a type of automated learning is a potential outcome," said Kawato. "However, in this study we confirmed the validity of our method only in visual perceptual learning. So we have to test if the method works in other types of learning in the future. At the same time, we have to be careful so that this method is not used in an unethical way."

At present, the decoded neurofeedback method might be used for various types of learning, including memory, motor and rehabilitation.



Source: National Science Foundation(NSF)