What’s better than coming in to work on a Monday to find a brand spanking new AI device on your teardown table? Finding two brand spanking new AI devices on your teardown table of course! And controversial devices at that!
On the one hand, we’ve had the drama around MKBHD being accused of attempting to sink both Fisker and Humane AI and on the other was the recent discovery that the Rabbit R1 was an Android app all along. And the question remains: Should these devices have just been an app?
Maybe. And maybe not. There’s an argument to be made against producing hardware that could end up as e-waste when a software solution is available. On the other hand, not everything in our lives needs to be tied to a smartphone, especially at a time when more and more of us are trying to reclaim our lives from the blasted things.
Regardless of how you feel about these two devices, they’re here now and, for us here at iFixit, that means we want to see what’s inside. For science of course.
Rabbit R1 Teardown
Gaining entry to the Rabbit R1 was a bit of a mystery at first. Is this like the older iPhones where you pop the screen off to gain entry or do you go in through the back plate? The back plate seemed to be the better bet and it paid off.
To my dismay, I didn’t find tiny bunnies running on hamster wheels to power the device. It’s a plain old lithium-polymer battery clocking in at 3.45 Wh. There’s room enough for a bigger battery in here and while it would increase the overall weight of the device, the bigger battery is definitely needed given how quickly the R1 tears through a charge.
The main housing contains a charmingly large motor and gear combo that powers the swivel mechanism on the R1’s camera. Attached to the end of this assembly is the camera sensor (of unknown resolution) and what appears to be a Hall Effect sensor that works in conjunction with the magnet inside the roller wheel (in the video we mistakenly referred to this as a rotary encoder. Oops!).
The mainboard houses the ICs that are running the show here…the local show at least. The camera, LCD panel, and small components are all run by this mainboard. Everything AI based is performed on the cloud. There simply isn’t enough processing power here for a local LAM or LLM to run.
Humane AI Pin Teardown
The Humane AI Pin posed a similar problem to the Rabbit R1, namely a new category of device where we might guess that a rear entry makes the most sense but we’re just not 100% sure. It turns out that the adhesive on the back of the Humane AI Pin is far easier to pry under.
The back cover is, perhaps unsurprisingly, essentially a charging coil wrapped in plastic. But what really stands out is the level of care that has gone into beautifying every surface of the interior. You can really see the ex-Apple engineer’s demanding need for perfection.
I won’t hash out the teardown here, we have the video for that, so let’s jump straight to the most interesting and innovative portion of the hardware: The MEMS laser projector. I pulled my Electrical Engineer colleague in for this one. Chayton Ritter explains: MEMS or micro-electromechanical systems combine electronic and moving parts to make all sorts of ultra-compact devices and sensors. The Humane Ai’s laser projector uses two piezoelectric MEMS mirrors in a galvo configuration to display information. A galvo configuration consecutively reflects light off two perpendicular mirrors that can rotate slightly. There are two mirrors for the X or Y dimension which change the laser location depending on their angle.
These mirrors are manufactured by STMicroelectronics and nicknamed Castor (Circular) and Pollux (Rectangular). Castor is a resonant mirror that constantly moves back and forth at a set frequency while Pollux is quasi-static and can be precisely positioned. To display an image, a laser first shines and reflects off the resonant mirror to create horizontal scanlines then reflects off the quasi-static mirror for the vertical rows. This all happens really fast so we can’t see the scanning and is similar to how images are generated on a CRT television.
Last but not least, we have the mainboard powering all the local hardware, with the LLM being based in the cloud and only accessible with a WiFi or cellular data connection. Again, these devices simply do not have the ability to run a localized version of their respective AIs when shrunk down to this size. At least not yet.
That’s it for this double teardown. As with any first generation device category, we don’t really have anything to compare the Rabbit R1 or Humane AI Pin to so we’re not going to score either of these devices. We hope that any future iteration of similar devices will provide easier access to their respective batteries.
But for now, let’s just admire the things that Humane and Rabbit did right and we’ll carefully watch and see how the hardware continues to evolve.
0 Comments