A 6-Legged Robot Stares at the Sky to Navigate Like a Desert Ant

In case you’ve been envying the desert ant Cataglyphis fortis lately, don’t. Skittering around the Sahara, the insect endures temperatures so brutal, it can sometimes only manage foraging runs of 15 minutes before it burns to death. Making matters worse, the heat obliterates the pheromone chemical trails that ants typically lay for each other to navigate. Get lost out here, and you’re literally cooked.

Accordingly, desert ants have evolved superpowers. They look for characteristic bands of polarized light emanating from the sun, which we humans can’t see, to get their bearings. They also count their steps to nail down a distance traveled, making them the fitness trackers of the insect world. Combining these two sources of information, the ants can zig-zag across the desert in search of delicious dead insects and still find their way home with remarkable accuracy.

Sensing polarized light is an indispensable skill for the ants, and perhaps soon it will also serve robots and autonomous cars. Researchers at the Aix-Marseille University in France report today in Science Robotics that they’ve engineered a six-legged robot, named AntBot, to find its way just like a desert ant. Not that your robocar of the future will navigate like this alone, but by leveraging polarized light, the machines could add a useful sense to augment fickle systems like GPS.

Because we can’t see polarized light from the sun, it can seem unintuitive to us paltry humans. Basically, it's a particular direction of propagation of the light. “Try to imagine there are lines in the sky oriented in a certain direction depending on the position of the sun,” says bioroboticist Stéphane Viollet, coauthor on the new paper. “There is a pattern in the sky, and this pattern is used by the ant to measure the heading.” It's like a massive map painted across the sky. As you can see in this handy video, filters can expose for the human eye what the ants see naturally.

To see like a desert ant, AntBot is using a surprisingly simple sensor, known as a celestial compass. It has two photodiodes that convert the sun’s polarized UV light into electrical signals. “This is absolutely non-conventional vision,” says study lead author Julien Dupeyroux. “These are very minimalistic sensors.”

LEARN MORE

The WIRED Guide to Robots

That’s the first bit of information the robot needs. The second is the distance traveled, which is straightforward: AntBot will also count its steps, just like its desert-dwelling muse. Ants also train part of their eye on the ground to get an idea of their speed, which is combined with the step count to give the critter an idea of how far it has traveled, and therefore how far it will need to amble to get back to the nest. AntBot does this as well with something called an optical flow sensor—basically, divining how quickly the ground is moving across the eye.

“You just need two pieces of basic information,” says Viollet. “You need your heading, and you need the distance traveled. When you decide to go back to your home, you can estimate your position with respect to the nest very easily.”

Ants have to be extremely precise with this kind of calculation, because there’s no room for error in the blazing desert. And it turns out that AntBot can also manage incredible accuracy, especially given how simple its sensing technique is. To test it, the researchers programmed the robot to “forage” much like a desert ant—that is, zig-zagging rather than going straight in one direction.

Take a look at the figure above. At left is an ant’s course, the thinner line being its outbound path and the thicker, straighter line being its return home. At right is the robot’s attempt (the solid points on the path are where it stopped to get its bearings). In outdoor experiments, AntBot managed to travel almost 50 feet, yet divine its way back to its starting point with an accuracy of less than half an inch.

The idea moving forward, then, is to adapt this system as a complement to other robotic senses, like traditional machine vision and lidar (which maps an environment by coating it in lasers). Both are computationally and energetically expensive, but AntBot’s sensors are much less intensive—remember, it’s just two pixels watching for UV polarized light. Plus, this kind of navigation works even when it’s overcast outside, because UV light can penetrate clouds.

It could also help compensate for the limitations of GPS, which are particularly problematic for self-driving cars. “There's a lot of metallic structures in cities, and this disturbs the magnetic field,” says Julien Serres, coauthor on the paper. “We think that adding this kind of visual sensor can help to get reliable information for the autopilot.”

For robotics more broadly, this approach is another example of how the natural world can offer design ideas to overcome the shortcomings of existing technology. Natural selection abhors a waste of energy—critters are optimized to generally use as little as possible as a matter of survival. Desert ants are no exception. What these researchers have done is co-opt a highly energy-efficient way of sensing the world, which they can then further refine.

“I think it's a strategy that works really well,” says Jeremy Fishel, cofounder and CTO of SynTouch, which has developed a system that gives robots the power of touch. “You study the biology, and then you bring it over to the artificial world, and then you can iterate very quickly.” Meaning, these researchers can take a system that natural selection has carefully honed over millennia, and further tune it for a robot.

So here’s to the desert ant, toiling in hell and inadvertently helping robots navigate this big, bad world of ours.