The U.S. Army may be able to replace its scout soldiers with robots that can discern threats through video cameras if scientists in Menlo Park are successful.
The Defense Advanced Research Projects Agency awarded SRI International a $3,068,155 contract for a project called Visual Intelligence Grounded in Learning, which focuses on developing technologies that would enable the automation of surveillance robots.
“Surveillance in dangerous areas is currently performed by remote reconnaissance teams and advanced scouting patrols who are often risking their lives," said Dr. Hung Bui, a senior computer scientist who is on the VIGIL team.
While their efforts can be supplemented through the use of cameras on unmanned vehicles, such devices still require constant human monitoring to analyze ongoing activities, Bui said.
“SRI’s VIGIL system combines automatic reasoning and machine perception to automatically analyze areas of interest while keeping members of the armed forces out of harm’s way,” Bui explained.
Researchers in the Artificial Intelligence Center of the Menlo Park-based research and development facility are working with the University of Leeds and the University of Maryland to find ways to consistently interpret behavior that could be threatening, based on known information about the actions of objects.
They have been working for the past nine months, and are still in the preliminary stages of the research, according to Dr. Bob Bolles, program director in the Artificial Intelligence Center.
The team has not deemed any specific objects such as shapes of people, or images of tanks, dangerous.
“The system at the moment is trying to describe what objects and actions occur in a video without performing threat assessments," Bolles said.
Members of the team, such as the principal investigator, have previously researched ways to create cognitive entities that organize data and learn, and may use this information when the time comes to fill in the gaps in percepetual input. That program was called the Perceptive Assistant that Learns, or PAL, for short.
“Our current approach is influenced by techniques developed under PAL, especially our use of probabilistic reasoning,” Bui said.
The DARPA contract is for two years and comes with priced options in the amount of $5,062,345.