Resilient Robots

Resilient Robots

It's one thing to teach robots to walk. But now, scientists have taught a robot to limp—and that's actually a big step forward.


Robots that compensate for injuries. I'm Bob Hirshon and this is Science Update.

If you sprained your ankle, you might walk with a limp or use crutches, but you'd find some way to get around. Now, scientists at Cornell University have built a robot that can do the same thing.

The team included computer scientist Josh Bongard, now at the University of Vermont. He says their four-legged robot actually learns how its body normally orients itself in space. If that suddenly changes, it checks itself for damage.

Josh Bongard:

After it's figured out what's gone wrong, it's able to come up with a new way of moving, or a new strategy, that allows it to carry on with its mission, despite the fact that it's been damaged or injured.

Next, they plan to program the robots to fix themselves, or even summon other robots for help. He says this resiliency could be useful in exploring other planets, or in surveying disaster areas here on Earth.

I'm Bob Hirshon, for AAAS, the science society.

Making Sense of the Research

Scientists are designing more and more robots to explore places that humans can't or shouldn't go, like the surface of Mars, the rim of an active volcano, or a dangerous disaster site. But what happens if those robots get broken or injured? Sending humans in to repair them would defeat the purpose of making the robots in the first place. And in some situations, it would be impossible anyway.

That's why Bongard's team wants to develop robots that can fix themselves. To do that, the robots will need to know when they're broken. They can only know that, and correct it, if they have some awareness of what's normal. For example, you can't recognize that you broke your leg unless you know how your leg feels when it's healthy and whole. In other words, these robots need a self-image.

That's not such an easy thing to program. Human babies and toddlers spend the first several years of their lives gradually understanding that they are separate from other people, that they have many different body parts and many kinds of physical and emotional feelings, and what it feels like to be sick or hurt. If it takes a brain as complex as ours that long to develop a self-image, how do you get a simple robot to do it?

You start by lowering your expectations. It's not necessary for a robot to think like a human, or refer to itself as “I,” in order to make it resilient. Nor does it even have to see, hear, or feel like we do. In fact, Bongard's team figured out that a robot needed only two sensors to diagnose and correct damage: one that detects whether the robot is tilting left or right, and another that detects whether it's tilting up or down. The information from those two sensors tells the robot how its body is oriented in space. From that, it can make inferences about possible damage. For example, if it attempts to stand so that its body is level, but actually tilts forward and to the right, it can realize that something must be wrong with its right front leg.

Still, the robot has to learn how to do this. It starts with what's pre-installed: the two tilt sensors and an internal virtual reality system that incorporates the basic laws of physics. While practicing various movements, the robot gathers information from the tilt sensors, and feeds the data into the internal virtual reality model of itself. Once the model has been fully developed, the robot can use it to track its position in space as it walks, runs, climbs, or crouches. When the information from its sensors conflicts with the model—for example, that its body should be level but in fact it's tilting backward—the robot infers that something's wrong, and can take actions to try and correct it.

As you heard, with some additional programming, it may be possible for the robots not only to compensate for injuries, but to fix them, or to communicate to nearby repair robots that they need help. Bongard says that their model might also be used to create robots that can respond to sudden changes in their environment—for example, to change their gait when the terrain changes from smooth to rocky, or to stop short when they reach a flooded area that could ruin the entire machine. That kind of self-sufficiency will be necessary for robots to take over dangerous human jobs.

Now try and answer these questions:

  1. What do the researchers mean when they call their robot “resilient"?
  2. In what sense is this robot self-aware? In what ways is it not self-aware?
  3. Are there kinds of damage that a robot could not detect with tilt sensors alone? Be specific.
  4. Consider your answer to #3. Suppose you had to make it possible for these robots to detect that kind of damage. What's the simplest kind of sensor or other equipment that you can imagine to do the job?

You may want to check out the December 8, 2006, Science Update Podcast to hear further information about this Science Update and the other programs for that week. This podcast's topics include: your birthday greetings to us, hopeful news about malaria in Africa, robots that can recover from injury, news about Neanderthals, the truth about lie detectors, and money brings out the best and the worst in us.

For Educators

In the National Geographic News article New Robot Reproduces on Its Own, you can learn about a self-reproducing robot and its possible applications.

In the PBS Interactive activity Bomb Squad: Hazardous Duty Robots, students are presented with three dangerous situations where hazardous duty robots could be used. They must read the scenarios, meet the robots, and then match the right robot to the job.

Related Resources

Night Lights
6-12 | Audio
Pushing Atoms
6-12 | Audio
Curve Balls
6-12 | Audio

Did you find this resource helpful?

Science Update Details

Grades Themes Project 2061 Benchmarks State Standards