ORLANDO — The vision in the 2004 summer blockbuster film iRobot was a world where humanoid robots work side by side with their masters.
Sometime in the next 90 years, science fiction will become science fact, some robotics experts predict.
Machines that walk upright will assist civilians and the military alike, said Stefan Schaal, associate professor of computer science and neuroscience at the University of Southern California.
“We should at some point be able to create an artificial human being and I think humanoid robots are currently the first step toward that,” he said at the Army Science conference.
“This is going to happen,” he predicted. “And it’s going to happen in this century.”
It may not be as “polished” as the iRobot movie, he added.
While other experts noted that there are huge technological hurdles to overcome, basic research continues on several critical technologies such as vision, movement and computational models that will allow robots to “think” like humans.
A parallel effort to map — or reverse engineer — the human brain is going to give robotics experts inspiration that will allow them to create these advanced models, researchers at the conference said.
The National Academy of Engineering is spearheading this “Grand Challenge.” Just as researchers successfully mapped the human genome earlier in the decade, the engineering community — not normally thought of as being a part of the life science discipline — says there will be a clear benefit to a Herculean effort to figure out exactly how the human mind works.
“If we could determine the software of the human brain, we could embed all sorts of systems so as to provide human like quality for machines,” said John Parmentola, director of research and laboratory management at the Army office of the deputy assistant secretary for research and technology.
Neural models will enable robots to better perceive, think, plan and act, said James Albus of the Krasnow Institute at George Mason University, Va.
“Significant economic and military applications will develop undoubtedly early in this century and in fact are already developing,” he said.
There are numerous other applications as well, including artificial retinas or ear implants for the visually or hearing impaired, and artificial limbs for the handicapped, said an academy fact sheet.
“Reverse engineering the brain is clearly one of the great challenges of the century and I believe there will be a lot of progress,” said Schaal.
Currently robots can be programmed to do simple tasks, but for a humanoid robot to work and live alongside people, they will be required to operate in the real world — not in tightly controlled laboratories or factories.
Computer modeling works fine for stationary robots working on an automobile assembly line by performing the same spot welding task over and over again. But put a robot outside and command it to walk from point A to point B — that is another matter.
You can’t model everything, Schaal said. The robot has to learn and adapt to myriad scenarios.
Schaal showed videos of robots in laboratories playing simple games such as air hockey and tennis. A backhand is relatively easy to program. But humans, once they learn the basic stroke, make slight adjustments in the speed or angle of the racket in order to hit the ball. The robot must learn some abstract mathematical equations that it can apply to real life.
There is progress being made in labs, as Schaal pointed out. The machines can be taught through reinforcement learning, better known as trial and error.
“It’s very similar to what we do with our children,” he added. “We teach them and then those children run their heads into walls and they get better over time and stop doing that.”
So how do you make a robot run?
The basic movement is simple enough. But there are rocks and curbs, wet surfaces and numerous other obstacles out there.
That will require robots to plan. Humans make the adjustments quickly. This task will require robots to perceive a problem and make an adjustment in a fraction of a second.
“The remaining tall pole in the tent is perception,” said Albus.
“Intelligent machines cannot achieve human levels of performance until they can perform as well as humans on visual tasks,” he said.
Modern sensors perform better than humans. They can see at night, in subtle shades of colors and farther away. Engineers can strap such sensors on to a robot, but they need the ability to perceive and understand situations and relationships, Albus said.
Yet this is not as large a technological hurdle as one would imagine, he maintained. For example, the number of objects a robot might encounter in order to walk down a road or a city street is not infinite, he said.
The numbers “are quite modest, at least from a computer science standpoint,” Albus said. There are roughly 10,000 objects that must be recognized. Other skills may require similar numbers. “For a computer, 10,000 things is not a big deal,” he added.
Schaal said there is little funding currently going toward teaching robots to learn on their own. The Defense Advanced Research Projects Agency’s “Little Dog” program is one exception.
Little Dog is a four-legged robot built by Boston Dynamics that is exploring how to move over rough terrain independently.
It is designed to “probe the fundamental relationships among motor learning, dynamic control, perception of the environment and rough terrain locomotion,” according to a company fact sheet.
Power remains another area that is receiving scant attention, the experts said. How will a robot mimicking human movement carry enough energy?
The technology has not advanced far enough out of the laboratory environment where it is a concern right now, Schaal said.
The actuators of the human muscles are unparalleled in terms of energy efficiency, he noted.
There are possible avenues to pursue such as passive spring elements that store and release energy as the robot moves, but currently no machine can come close to the performance of a human or animal muscle.
The Little Dog, measuring less than a foot in length, has about 30 minutes worth of battery power, the company fact sheet said.
Human muscles are “far beyond anything that can even be conceived right now,” Albus said.
As far as humanoid robots, the United States is a distant third behind Japan and Europe when it comes to advancing the technology, the experts said.
The Grand Challenge to reverse engineer the human brain may accelerate the pursuit of assistive humanoid robots that work alongside their flesh and blood counterparts, Albus said.
“We’re at a tipping point analogous to where nuclear physics was in 1908,” he said.
Supercomputers are approaching the computational power of the human brain. Neural models that will emerge from the Grand Challenge will enable perception, thinking, planning and acting in computers.
Human level performance may be feasible on desktop class computers within two decades, Albus predicted. “That would really change everything.”
“Technology is emerging to conduct definitive experiments — not just philosophical conversations,” he added.
J. W. Singer, a Brookings Institute scholar and author of “Wired for War: The Robotics Revolution and Conflict in the 21st Century,” cited a survey of military robotics scientists and officers involved in developing the technology that asked when they believed humanoid robotic infantrymen would appear on the battlefield. The scientists’ average response was 2020. The officers said about 2025.
He predicted that when they do appear, the machines will probably not be replacing real soldiers or used to fire weapons. It will be more of an assistive relationship akin to a law enforcement officer and his police dog.
“Unmanned systems do what they do best, and humans do what they do best — working together. That seems to be where the latest war gaming is taking us,” he said at a Brookings panel discussion in Washington.
Another larger question that remains to be answered is the cumbersome way the U.S. military develops and procures weapon systems. Acquisition normally takes years of development, testing and evaluation before any equipment makes it to the battlefield, he noted.
“Is that the right industrial model for this future of [robotics] war?” Singer asked.
The implications for the military are profound, Albus noted.
Intelligent weapon systems with human level capabilities will outperform manned systems because they don’t have to eat, sleep or go to the bathroom.
They will cost less to train. “Train one or two and you can download a CD to the others,” he said.
They can be stored in a box and placed on a cargo ship like any other weapon — only taken out when needed. And most importantly, they can keep soldiers out of harm’s way, he added.
“My feeling is that intelligent systems will revolutionize warfare in many, many ways,” Albus said.
Of course, in the movie iRobot, the machines went wild and started killing everybody, Schaal joked. Hopefully, that will remain science fiction and not science fact.