The trouble with programming robots to walk over shifting terrain like a desert is that what works in one environment may not fly in another. These conditions are often inconsistent, with rocks and sand of all shapes and sizes to contend with. One way around that would be to have the robot teach itself, which is exactly the approach researchers at Arizona State University have taken with their so-called C-Turtle, a cheap, nature-inspired robot that could one day be used to clear landmines and explore Mars.

Robots that learn as they go are getting better all the time. We have seen robots learn to cook by watching YouTube videos and others that teach themselves to grasp objects like a human. And if those aren't quite stoking your Skynet fears, then how about a "mother robot" built by UK scientists that can not only build its own child robots, but mimic the process of natural selection to improve the capabilities with each generation?


Read the site and newsletter without ads. Use the coupon code EOFY before June 30 for 30% off the usual price.


The idea behind making the C-Turtle self-learning was so that it can learn to tackle the shifting sands of various desert-like terrain. Doing this required minds from the university's computer science department, along with researchers in mechanical engineering and biology.

The result was an inexpensive robot made from thin cardboard, using motors that cost around US$5 and chips around $10. Each robot overall costs around $70 and features a curved leading edge and curved flippers, just like a turtle, along with learning algorithms to help it scuttle across the sand.

"It turns out the ones shaped like that work better than just a square paddle," says team member Andrew Janssen. "We tested things that are impossible in nature. They didn't work."

The team put the C-Turtle to work in Arizona's Papago Park, where it took around one hour to teach itself to walk across the terrain. It did this by digging quite hard to propel itself forward, but not hard enough to dig holes and get stuck.

With further work, the team imagines these robots could one day be deployed in packs, as a form of swarm robotics, to monitor conditions or clear minefields. And possibly one day, stacks of cardboard and a laser cutter could be sent to Mars, allowing fleets of C-Turtles to self-assemble and explore the Martian landscape.

The team will present two papers on the robot at MIT and Stanford this (northern) summer, describing both its biological inspiration and the learning algorithms behind it. You can see the C-Turtle in action in the video below.

Source: Arizona State University

View gallery - 3 images