Why we did not build a robot pet, after buying one
We did not build one of those cute, non-speaking companion pets, because I bought one, held it, and concluded that reaching the standard where it fools your hand is beyond what we can currently do.
What holding one tells you
Several domestic suppliers make this category. Watching videos before buying, it looked genuinely soothing. A few strokes in, I knew it was not going to fool my hand.
You can clearly feel the motor vibrating inside, and the shell moves on a fixed rhythm. It is not alive. Anyone who has held a real cat knows there is a randomness to it, the force of a paw curling, the resistance of fur, an occasional twitch of a leg, and none of that reproduces from a program easily. The closer a robot pet gets to it, the more it costs.
Three layers have to land at once
Sensor density
One pressure sensor is not enough. There are many ways to hold a thing, with different force in different places, and the reaction should differ accordingly. That requires a distributed sensing layer, not a switch.
Motion fidelity
The motor and mechanism have to produce movement that feels elastic and slightly unpredictable, different each time. That is a demanding mechanical design problem.
Material and fur process
The surface has to read as neither fabric nor silicone but something biological. Sampling cycles here are long and tuning eats time.
Each layer costs money on its own. Together they cost considerably more, and none of them is the kind of problem where spending guarantees the result. They need repeated sampling and repeated validation until the feel genuinely passes.
Why we closed the road
I try to be clear-eyed about the edge of my own capability. On sensing, mechanism and materials we do not have deep enough supply-chain experience, and we do not have the time for a dozen rounds of sampling.
More importantly, the entry bar for this category is that the feel is convincing. Below about ninety out of a hundred, the product is worse than not existing, because the user has held a real animal. Touching something obviously mechanical after that is a bigger letdown than touching nothing. This is not a category that tolerates a partial build.
Rather than force out a version that does not qualify, we went down a road we can actually do well.
The demand is real, the mechanism is not ours
The underlying need is genuine. Companionship without litter trays is a real requirement for people who rent, who are busy, who are allergic. But that need is redeemed through touch, not through speech. We cannot do touch to the standard it requires, so we spent the effort where we can deliver, which is voice, knowledge and companionship through conversation.
The hardest step in product judgement
It is not coming up with a good idea. It is recognising which road you cannot walk, and then closing it.
Closing it means admitting you are giving up a market and a very attractive story. But shipping below the standard a category demands damages what users expect from the whole category, including from the people who could have done it properly.
Related
Want to try the hardware behind these notes?
Six AI hardware platforms are in mass production. Tell us your market and channel and we send an evaluation sample.
Request a sample