Oregon State University has received nearly $9 million to lead a four-year research project aimed at infusing artificial intelligence and robotic systems with more common sense.
Machine-learning researcher Alan Fern of the OSU College of Engineering and two collaborators will develop and train a “machine common sense service” that will learn about its environment in a manner similar to that of a toddler.
Fern is working with roboticist Tucker Hermans of the University of Utah and behavioral psychologist Karen Adolph of New York University on the $8,741,152 project funded by the Defense Advanced Research Projects Agency.
“We are studying and developing learning and reasoning techniques to enable AI systems to exhibit common-sense reasoning and planning capabilities on par with those of an 18-month child,” Fern said. “A key aspect of our approach is to study how to effectively combine the representation-learning capabilities of deep neural networks with the powerful reasoning capabilities of state-of-the-art AI planning and reasoning engines.”
Representation learning, also called feature learning, is the means by which intelligent systems gain and categorize information, such as information about places and objects in their environment.
Using video of toddlers provided by Adolph, Fern will make a computer model of how babies explore their environment and then create an “artificial agent” – a virtual toddler – that will be tested in a simulated 3-D environment.
“It will look like a simple robot in a video game exploring a virtual space,” Fern said. “The basic idea is to get robots to have more common sense regarding physical interactions in their environment.”
Source: Oregon State University