Software helps robots make intelligent decisions

Rapidly changing situations require robot sensors to be smarter

In its quest for ever-smarter robots, military researchers this week gave a $850,000 grant robotics and automation vendor TRACLabs to build software that can help robots use their sensor systems more intelligently.

The Defense Advanced Research Projects Agency (DARPA)-funded project will develop software that in its first stage, will help merge the many data streams of robot sensors to improve their recognition of static objects, such as bomb components or helping the bots perform tasks such as opening doors and drawers, TRACLabs stated.  In the second stage, the focus will shift to searching for objects that are able to move, from mobile weapon units to human targets.

According to DARPA, the number of sensors mounted on unmanned vehicles, satellites, surveillance cameras, etc. continues to grow. These sensors offer a wealth of data, but converting these sensors data into a coherent, symbolic view quickly is a trial.

The problem of mapping sensory input into discrete categories is a fundamental challenge that must be addressed by any advanced sensor-processing system. The simplest of gauges map physical measurements into human-accessible categories, like "speed," "temperature," "RPM" and other common measurements, DARPA stated.  In more complex sensor-processing, like automatic image analysis, the challenge of converting the many low-level bits of information into just a few user-relevant categories is greater, DARPA stated.

When existing categories are insufficient and need to be extended, refined, or redefined, the situation must first be identified by people, usually end users or soldiers and then resolved by engineers in a process that can take months or even years. In adversarial situations, when opponents are actively seeking to modify and disguise the signatures of their equipment, the ability to recognize novel patterns in sensor data and reason about the cause and threat potential can be a decisive advantage. As the number and type of sensors continues to grow, so does the need for adaptive algorithms that can learn the appropriate categories and categorical distinctions required by the end users and applications, DARPA stated.

That's where the TRACLabs software comes in.  "Imagine a situation where a unit wants to see if chemical or radioactive weapons have been deployed in a building before they enter. Our software will enable a robot to quickly and accurately assess the situation without putting troops at risk of exposure," said David Kortekamp, President and CEO of TRACLabs in a release.

DARPA uses another example:  A thermal-imagery based sensor processing system could be capable of classifying various vehicle signatures. Assuming that it classified Humvees as friendly military vehicles, such a system may be confused by the altered signature presented by Hummers in civilian areas. A sensor intelligence might be able to first note the discrepancy, then assess the possible explanations, potentially in a dialog with users, and then apply standard machine-learning techniques in an attempt to produce an accurate assessment of a Hummer as a civilian vehicle.

If guerilla forces then begin to use Hummers for troop movements, the more intelligent system might be able to use information about location, the structure of convoys, and common operating knowledge to produce an updated discrimination between the non-threatening use of Hummers and potential adversarial uses. This effort will create a foundation for the construction of robust, user-centric sensor-processing systems that integrate large and diverse sets of advanced information-processing capabilities, DARPA stated.

Such robotic intelligence is being developed under DARPA's Architectures for Cognitive Information Processing (ACIP) program. DARPA's cognitive computing research is developing technologies that will enable computer systems to learn, reason and apply knowledge gained through experience, and respond intelligently to new and unforeseen events.

Another ACIP project is artificial intelligence software known as a Machine Reading Program (MRP) that can capture knowledge from naturally occurring text and transform it into the formal representations used by AI reasoning systems. The idea is that such an intelligent learning system would unleash a wide variety of new AI applications - military and civilian -- ranging from intelligent bots to personal tutors according to DARPA.

In other robot-enhancing activities intelligent robot vendor iRobot last year licensed Laser Radar or Ladar technology for use in its line of military robots, a move that could result in a new line of machines that can see and operate more effectively in dangerous situations.  Such small, advanced robots could be deployed in less than a year, experts said. 

Specifically the robot-maker is licensing Advanced Scientific Concepts' 3-D flash Ladar which uses laser beams to scan and process targets. The system has the ability to create a virtual 3D picture of an entire area. 

Layer 8 in a box

Check out these other hot stories:

"Faux Predator" flies to train unmanned aircraft pilots

High-capacity satellite system zips voice, data, video to aircraft

Military looking for ways to spin seawater into fuel

Sears forced to destroy spyware-gathered information

NASA's future manned spaceflight forecast: Bleak

Military spends $10M to build Web sites aimed at squelching anti-US messages

NASA teams to build pod-like, high-speed everyday transport system

Mach 6 test aircraft set for trials

Has US been unfairly torched over the electronic waste disposal issue?

Deep-space clouds smell like mothballs?

Join the Network World communities on Facebook and LinkedIn to comment on topics that are top of mind.

Copyright © 2009 IDG Communications, Inc.

SD-WAN buyers guide: Key questions to ask vendors (and yourself)