“There are very good walking systems out there that are extremely energy-efficient. However, those aren’t designed to carry payloads or move over rugged terrain. Then you have systems like Big Dog that can walk over rugged terrain, but not very efficiently.

What we need to do is combine the two. Once we do, our power requirements are much lower and we might get into the capabilities of fuel cells. So by using control to help us address power requirements, at some point, we will have advanced the state of the art in fuel cells to the point that we can produce much more animal-like vehicles in terms of endurance and the ability to move through rugged terrain.”

Other DARPA program examples include LANdroids (Local Area Network Droids), which is entering into its second phase. The concept is to ensure radio communications when a squad storms a building. Each foot soldier carries several LANdroids, tossing them out of his bag as he moves through the building. They are basically mobile radio nodes, tasked to configure themselves into an ad hoc radio network, then reconfiguring to compensate for the loss of any node destroyed by an enemy, providing every soldier with assured high bandwidth to communicate voice, data, and video back to a base station.

In terms of manipulation, the most advanced of currently fielded robots are those used to defuse or destroy IEDs. Hundreds of those have been deployed to Southwest Asia, where the number of IEDs they had dealt with through the end of 2008 exceeded 11,000. While there is no way to determine how many lives those robots saved, it easily could average two or more per IED. As successful as that effort has been, however, those robots still require a human controller.

Practical autonomous manipulation is as much a matter of perception as degrees of physical movement and fine control. But perception, from depth to recognition, requires far more than a good imaging system. Major breakthroughs are needed in cognition to understand images, for the robot to be able to interpret its sensors and fully understand what is going on around it, for navigation as well as manipulation.

“There is a difference between seeing and understanding – cameras see, they grab an image, but human eyes understand what they see within a split second of opening,” Mandelbaum said. “But a human’s full understanding of separating and identifying objects and their location is still a ways off in terms of computation and robotics. We’re getting there and there has been a tremendous amount of advancement in the last 20 years, but we still have a very long way to go.”

In 2000, DARPA set out to advance robotic technology in terms of mobility and perception, initially through two parallel programs that decoupled those issues. The Unmanned Ground Combat Vehicle (UGCV) looked at a high-capability platform with no perception, while Perceptor used standard all-terrain vehicles (ATVs) to develop sensor packages and algorithms to do sensor processing and planning for off-road navigation. When completed in 2004, DARPA took the best performers of both and put them together into one system – UPI (UGCV Perceptor Integration).

“UPI was a very capable system, working in complex terrain, but it was not designed to work in and among people or maneuver among things moving about it,” Mandelbaum said. “On the other side, you have Urban Challenge, which demonstrated the ability to work in and among other moving vehicles and people in a structured terrain of roadworks.”

DARPA began its Challenge competitions in 2004, offering a $2 million prize to the first competitor (typically academic or corporate teams) able to complete a 132-mile course across the California desert in less than 10 hours with no human intervention. Some teams used highly modified commercial vehicles; others essentially built their own. In 2005, a team from Stanford University became the first to meet all requirements of the Grand Challenge.

That led DARPA to up the ante with Urban Challenge – not only did the vehicles have to duplicate the original challenge capabilities, they also then had to maneuver through urban traffic, both other vehicles and pedestrians (mannequins) set up on an abandoned California military base, without incident and without violating standard traffic laws. A team from Carnegie Mellon University completed the course and met all requirements to win first prize, but five other entrants also crossed the finish line.