from Technology & Learning

The next generation of robots has evolutionary capabilities.

Robots that can learn new behaviors. Robots that can reproduce themselves. Science fiction? Not anymore. Roboticists at Cornell's Computational Synthesis Lab have developed just such engineered creatures that offer interesting implications for education.

Traditionally, every move a robot might make is preprogrammed by its creators. Their ability to complete tasks is only as good as their inventors' ability to foresee every possible maneuver they might need to make and include it in the robot's job description. As a consequence, current robots can only help us with mundane, repetitive actions.

The team, headed by Hod Lipson, was intrigued by the question, "How can you get robots to be creative?" To answer it, the group set about making robots a different way. Instead of designing the machines manually with programs for every eventuality, the researchers created learning systems that would enable the robots to learn and evolve, more as children do.

Using this evolutionary robotics approach, the researchers created simple, legged robots with eight motors and some basic circuitry to allow them to try out various behaviors and act on the consequences.

We're not talking C3PO here, or even R2-D2. The appearance of these robots is not anthropomorphized. In fact, they look more like four-legged tarantulas made out of LEGO toys. But they do mimic humans in their ability to learn about themselves.

Through self-directed interaction with their environment, the robots develop a sense of their own shape and structure. After flailing about for a short time, they learn to move forward. Even more amazing, lop off one of their legs and they figure out how to keep moving along using the three remaining ones.

These robots may be brainy, but they haven't yet figured out how to reproduce themselves. That feat, however, has been accomplished by some of their robot cousins, which were also created at Cornell's Lab.

At rest, one of these reproducing robots looks like a simple stack of toy blocks. But once it goes into action, it's a different story. Performing a fascinating swivel dance, the robot collects additional modules from a nearby supply and constructs a replica of itself. As long as the researchers continue to supply the raw materials, the robot can continue to reproduce itself, as can its offspring.

While these robots are not quite ready for jobs in the classroom, the work of developing them may benefit educators. "How do you decide what problems to propose to a learning system so that it can gradually learn something?" asks Cornell's Lipson. "How do you generate incremental, successive problems such that each one is learnable?" Those are essential questions for educators as well as computational biologists.

Want to learn more? Visit the lab's Web site.

A former teacher and principal, Michael Simkins directs TICAL, a California educational technology service.