Surveillance robots know when to hide

• 21 March 2011 by David Hambling
• Magazine issue 2804. Subscribe and save
• For similar stories, visit the Weapons Technology and Robots Topic Guides

The creation of robots that can hide from humans while spying on them brings autonomous spy machines one step closer

THE spy approaches the target building under cover of darkness, taking a zigzag path to avoid well-lit areas and sentries. He selects a handy vantage point next to a dumpster, taking cover behind it when he hears the footsteps of an unseen guard. Once the coast is clear, he is on the move again - trundling along on four small wheels.

This is no human spy but a machine, a prototype in the emerging field of covert robotics. It was being put through its paces at a demonstration late last year by Lockheed Martin's Advanced Technology Laboratories at Cherry Hill, New Jersey. With an aerial drone to their credit (see "Unseen watcher in the sky"), the company now wants to design autonomous robots that can operate around humans without being detected.

What makes the robot special is its ability to build a computer model of its surroundings, incorporating information on lines of sight. The robot is fitted with a laser scanner to allow it to covertly map its environment in 3D. It also has a set of acoustic sensors which it uses to distinguish nearby footsteps and their direction.

Lead engineer Brian Satterfield says the robot was designed to operate within four constraints: "Avoiding visible detection by sentries of known locations, avoiding potential detection by sentries whose positions were unknown, avoiding areas in which the robot would have no means of escape, and, as this robot was designed to run at night, avoiding areas that were well lit." To make it hard to spot in the dark, the robot was painted black.

If the robot believes it is in danger of being detected by an approaching sentry, it will try to get to a place where it can hide, Satterfield says. His comment is an example of how natural it is for us to talk about such robots as if they understand how they are perceived and have a "theory of mind".

"Lockheed Martin's approach does include a sort of basic theory of mind, in the sense that the robot makes assumptions about how to act covertly in the presence of humans," says Alan Wagner of the Georgia Institute of Technology in Atlanta, who works on artificial intelligence and robot deception.

But the level at which the robot's software operates is probably limited to task-specific instructions such as, "if you hear a noise, scurry to the nearest dark corner", he says. That's not sophisticated enough to hide from humans in varied environments.

"Significant AI will be needed to develop a robot which can act covertly in a general setting," Wagner says. "The robot will need to consider its own shape and size, to have the ability to navigate potential paths, [to be aware of] each person's individual line of view, the impact that its movement will have on the environment, and so on."

Satterfield's robot was built with off-the-shelf components. Both he and Wagner say that specialised hardware which is more compact and quieter will improve future robots' mobility and their ability to stay hidden. "There are very few fundamental limits that would prevent robots from eventually conducting extended covert missions and evading detection by humans," Satterfield says.

Lockheed Martin's work looks ready to emerge, albeit quietly, into the real world. The US army recently solicited proposals for a "persistent surveillance" robot with concealment capabilities and suited for extended deployments. Later this year, the US Department of Defense is expected to back that up with cash awards for working designs.

Air-Blowing Robot Makes Ping-Pong Balls Jump Through Hoops

Sure, you can make a ping-pong ball float just by blowing at it through a drinking-straw, but wouldn’t a ball-levitating robot be so much more fun? Luckily for us, University of Illinois grad students Aaron Becker and Robert Sandheinrich answered “yes” to this question, and built this incredible machine:

It’s called the Robo-Air Blower, and while the principle is pretty simple, the physics behind it are complex. A gimbaled nozzle fires compressed air at 620 kPa of pressure. This jet creates a fast-moving, low pressure area around the ball, trapping it. The jet is powerful enough to lift balls of between 24mm and 194mm in diameter, and up to 188-grams of mass.

But the tricky part is control. Fluid dynamics are a chaotic thing, and the programming of the robot control multiple balls, as well as non-spherical objects, like the water-bottle in the video, is complex. The robot’s brain is fed by two stereo cameras which track the balls’ movements and adjust the jet based on an algorithm.

Despite this somewhat dry explanation, the results are spectacular. The robot can push the balls sideways and diagonally, and make them jump through hoops. It looks like some kind of iPhone or Android video-game brought to life, or an up-to-date version of the old loop-and-live-wire game we played in school.

Robots learn to walk like a senior citizen

Today's humanoid robots are able to run, somersault and even dance – now comes a robot that walks like a senior citizen. It leans on objects in its environment for support to help it move around and complete tasks.

Robots, and more importantly roboticists, are looking at objects in the wrong way, thinks Sébastien Lengagne of Japan's National Institute of Advanced Industrial Science and Technology in Tsukuba. "Roboticists usually just see objects as obstacles to be avoided," he says "But they can help us."

Lengagne and his colleagues are developing a system to allow humanoid robots to use their entire bodies, and any surrounding objects, to help them move around cluttered environments and complete complex balancing tasks without getting stuck or falling over.

"If I ask you to look below your desktop, you will put your hand on the desktop for support," he says. "But most methods will try to get the robot to do the task without touching the desktop."

The team's robot, HRP-2, acts more like a human. It will place both arms on a table to maintain its balance when trying to sit down in a chair, or use one arm for support when taking a big swinging kick at a ball.

Sidlebot

The system breaks down tasks into two stages. In the first stage, software developed by Lengagne's colleague Karim Bouyarmane identifies objects in the robot's surroundings that it can use to help complete a task – for instance, leaning on a table with its forearms to sidle past it and into a nearby chair. The software then calculates a number of poses that the robot could strike to make best use of the table for stability while it shuffles towards the chair and sits on it. Lengagne's software then converts these static poses into one smooth motion, taking into account the forces operating on the robot in each position to ensure it does not lose balance.

At present, the system has to run these calculations on an external computer, but the team hope that ultimately a robot's onboard computer will be able to carry out the process in one step.

The work was presented earlier this month at the Humanoids 2010 conference in Nashville, Tennessee.