The Clever Hack That Let A Toymaker Bring Robot Cars To Market 20 Years Early

Anki Drive uses technology similar to Google’s self-driving cars, at 1/1000th the cost. Not bad for a toy car.

What is “cool?” If you’re a kid, cool is robot cars with lasers and machine guns, blasting around a track at 250 miles per hour. If you’re an adult, cool might be sitting back and reading a book or watching a movie as your Google self-driving car drives itself down the highway. In both cases, it’s the car that’s cool, not the dumb old road underneath.


But sometimes “cool” can be deceiving. In the case of Anki Drive–the iPhone-controlled, deathmatching cars Apple first revealed on-stage at the Worldwide Developers Conference back in June–the track itself is just as cool as the cars. In fact, that big plastic track that ships with Anki’s robot racing set is a brilliant hack that has brought the technology behind driverless cars to a mainstream consumer audience decades before its time.

If you’ve never heard of Anki Drive, chances are that it’s the kind of toy you would have gnawed off your own arm to have as a kid. After buying the Anki Drive starter kit, you unroll a Tron-like plastic track on the floor, load up the official Anki Drive app, pair the two enclosed Anki sportsters to your iPhone via Bluetooth, and give them a gentle push to start racing down the track.

But while Anki’s cars will happily drive themselves on a high-speed excursion around the track until they run out of battery, things get really cool once you take over control of a car with your iPhone and begin racing against a friend. Use your iPhone’s accelerometer to smash into another car, knocking it off the track, or use your onboard weapons to shoot the other car down with lasers and tractor beams. It’s awesome fun–a physical mash-up between Hot Wheels and Mario Kart–but what’s really cool is how it all works.

“There are three major types of problems engineers have to solve to make self-driving cars a reality,” says Anki cofounder Boris Sofman. “Positioning, reasoning, and execution.”

In other words, for a car to drive itself, first it needs to gather data about where it is in the real world: in human terms, it needs to be able to “see” where it is. Next, a car needs to be able to take this data and crunch it into a mental model of everything it can see and where everything is in relation to where it wants to go, then come up with a plan to get there. Let’s call this “understanding.” Finally, a self-driving car needs be able to execute a plan based upon its understanding of both its goals and environment by being able to control its own movement.

In many ways, this last part is the simplest challenge when it comes to making self-driving cars a reality. More complicated–and, importantly, much more expensive–is teaching a car not to just see, but to understand. Google’s driverless car uses upwards of $150,000 worth of equipment to see the road around it, including a $70,000 64-beam laser radar system. Anki’s solution is far simpler. “The track itself is the layer that bridges the physical and virtual worlds of Anki Drive,” says Sofman.


Although the surface of the Anki Drive track might appear an impenetrable, vinyl-like black, it’s actually transparent to infrared light. In fact, it’s made of the same plastic that lets the infrared beam shoot out to your TV inside your universal remote. Strip this layer of plastic away and what you’d see is an elaborate tarmac of code that each Anki car can see with the infrared eye of a cheap, downward-facing sensor. By reading this river of invisible ones-and-zeros, each Anki car knows exactly where it is on the track at any given moment.

After reading this information from the track, each Anki Drive car beams what it “sees” to a connected iPhone, which combines this information with the data from the other cars on the track into a virtual map it understands. The app then accounts for things like momentum and trajectory, then sends commands back to the cars on the track, telling them to speed up, slow down, or veer left and right, depending on both where the car is supposed to be going and what a player wants each car to be doing at that moment. Rinse and repeat, up to 500 times per second, and you have Anki Drive.

Anki’s solution to a complicated and expensive problem is an elegant one, but unfortunately, it can’t be rolled out on a nationwide scale as easily as simply unrolling a plastic mat. Without a centralized computer keeping track of every car in the country and rerouting them accordingly at all times, Anki’s approach–encoding our roads with infrared data–isn’t going to get driverless cars here any quicker.

Even so, Anki Drive shouldn’t just be dismissed as a toy. The way Anki cars see the road around them may differ, but that’s just the input. “Anki cars drive themselves using the same class of algorithms that Google uses for its fleet of driverless cars,” says Sofman. Just like GPS, radars, or lasers, Anki’s track is just another way of seeing. It’s teaching a car to understand what it’s seeing that is the hard part, and by going cheap and simple, the holiday season’s hottest toy may have just jumped self-driving cars to market by 10 or 20 years. What’s cooler than that?