Back in January 2014 I purchased an Anki Drive "robot car" setup, along with an iPad to control it. I had high hopes. Anki had a great story: originated in CMU's robotics program; raised $50M in VC funding (now more than $105M); and launched at Apple's WWDC conference in 2013. It was touted as "bringing robots out of the lab and into homes, using real AI." I couldn't wait to get my hands on one -- to try it out and to tear it apart to figure out how it works. Read on for details, including an inside look at how the robot localization works. But spoiler alert... I was disappointed.
This is Anki Drive, premiering on-stage at WWDC 2013 :
iFixIt did a beautiful teardown of the car (only), so I won't repeat that exercise. It's all pretty straight forward. A 32-bit ARM Cortex-M0, a Nordic BLE chip for comms, a lithium battery, and a basic DC motor drive train. The little optical assembly (bottom left picture) provides the biggest clue to how the system works: it contains a pair of infrared emitters and a lens assembly backed up to a little optical imaging sensor -- a lot like a higher-resolution optical mouse.
There is surprisingly little online information about Anki's special track. I took a few closeups of the mat (below, left and center). But these don't really show how it works; the only photo that kinda-sorta showed how it works is shown below (right), via FastCompany.
Unacceptable. :) It was obvious that the track had an IR-transparent coating, so I purchased a PublicLab Infragram webcam (approx $55 in their store) for my home lab. It's basically a stock webcam with a blue filter (Rosco #2007 or Schott BG3) that allows near infrared to be captured in the camera's red channel.
Looking at the track in infrared makes it pretty clear how the car works: It's glorified line following with patterns of dashes (varying width and darkness) alongside the line to help the car identify its lateral position on the track. Not exactly the "cutting edge technology" I imagined from all the media hype.
At the tail end of 2013, with a high-paying job on the horizon (I was slated to start at Google[x] in early January), I figured, "What the hell...." I went out and spent $600 at the Apple store to buy an Anki ($200) and an iPad ($400) to control it. At about the same time, I purchased the $55 IR camera; I'd wanted one of those for my lab anyway, but this was a good excuse to buy one. That brought the total up to ~$700 for this little experiment (accounting for taxes). As my wife reminds me: That's a pretty hefty price tag for what amounts to little more than a novelty.
The initial unboxing was pretty impressive. Anki spent a lot of time and energy to make sure the product and packaging looked good. No wonder Apple was keen to bring them onto iOS. But there was a hitch: The cheap version of the iPad didn't support BLE (bluetooth low energy, required by Anki). This wasn't really Anki's fault; I had even asked the "Genius" at the Apple Store if the iPad I purchased had BLE and if it was compatible with Anki... he said, "yes." Tool. A week later, I made another trip to the store, plunked down another $100 for a better iPad, and now all the hardware was in place.
I had promised a few friends (John O'Hollaren from Apple and Evan Ackerman from IEEE Spectrum) that I'd wait for them to play it for the first time. We all met up on Super Bowl Sunday in early February. Our halftime activity: battling on Anki!
Even with just two cars, it was fun. We played against each other for the duration of the 20-minute halftime, and then for another 15-minutes or so into the 3rd quarter. But by then, we were basically over it. We went back to watching football and never returned to the Anki except to put it away.
I played Anki just two more times in the ensuing 10 months: Once with a bunch of robotics enthusiasts up at Google[x], and another time at a party with friends who had been wanting to try it out. The effect was exactly the same: People would play for 15-30 minutes and then get bored -- classic XMas toy syndrome. And another big drawback: Anki is big! It takes up an entire room (a precious commodity in SFBA). Your only viable option is to setup and teardown after each use -- and overcoming that sort of activation energy for 15 minutes of gameplay is just untenable.
So the long and skinny of it? Anki is basically a pair of glorified line following robots combined with basic video game AI on the phone; not the revolutionary advances I was expecting. It's got some "cool factor" in concept, and amazing execution (technical and marketing)! But it's difficult to overcome the activation energy to setup and teardown. And even once you do get everything setup, the toy loses it's appeal after less than an hour.
I bears mentioning: Perhaps my view is a bit skewed; I may not be their target market. I'm not a young kid. And I literally bought an iPad to play with Anki, so the price was massively inflated. I hear that they recently upgraded their design to make it compatible with both Android (via WiFi) and iOS, and dropped the price to $150, which is much more palatable when you already own a controller. But frankly, I'd be happier with big 'ole RC car (like this Traxxas RC car), a 3D Robotics drone, or even a little micro quadrotor. I've gotten hundreds of hours of enjoyment out of these "robot toys."
It's a bit presumptuous on my part, but I don't believe that Anki sold anywhere near as many devices as they'd intended. If I were them, I would use whatever remains of that $105M in venture money to pivot (quick!). More tracks and cheaper cars won't dig them out of this hole.
After such a glowing review.... I have something to sell you: my Anki Drive is currently listed for sale on eBay. ;-)