4 ms·

This raises so many questions: 1. How many pictures do you need? It seems they keep sending out pictures and then they send more people to get more pictures.

This raises so many questions: 1. How many pictures do you need? It seems they keep sending out pictures and then they send more people to get more pictures. Is there a lot of additional value in additional pictures? 2. How hard is it to reverse engineer tappy? Once you know how it works, I just assumed that Huawei has more than enough engineering talent to put together a tappy. 3. Was there a particular component of tappy that was hard to copy? 4. If I had to guess, the software that recognizes if the phone screen is on and working is the secret sauce. Not the mechanical tapping which, although sophisticated, is presumably copyable.

jacobush 8y ago
I tried to build something very much like "Tappy" as an internal testing tool, and with the limited time and effort I put in, #4 was definitely what prompted me to give up and go for a "screen capture over ADB" instead of a camera. There was glare, all kinds of distortion, noise etc, that while of course possible to overcome, was way out of my league to deal with. I guess one could have thrown machine learning in the mix too... :-P
ethbro 8y ago
In answer to #1, I'd assume because Chinese PMs are much the same as PMs elsewhere. PM: "Why aren't we done yet?" Dev: "Umm, we need more photos." {Cue never ending stream of request for photos}
sonnyblarney 8y ago
Physical things are hard, there are a million little issues. Sometimes 'seeing' a piece or component can reveal a lot as it represents the solution to a problem that is not otherwise evident. The smallest details like the size/type of motors, power ratings, how they are positioned, what kinds of sensors they use can reveal a lot. To boot, you're not dealing with a toy that can fall apart after 10 uses - this is gear that ostensibly has to be fairly robust, which introduces a whole other set of quality challenges beyond 'just making it work'. Example: making a little 'toy car' that ran across the floor, we ran into all sorts of control issues with the motors as the car encountered surfaces of different types, and also issues with getting caught on carpet etc.. Upon seeing how our partners made a previous device, it became pretty clear how they solved the problem. Instead of testing various different configurations and trying to find the 'best fit' - we just used their example.

jacobush 8y ago

I tried to build something very much like "Tappy" as an internal testing tool, and with the limited time and effort I put in, #4 was definitely what prompted me to give up and go for a "screen capture over ADB" instead of a camera. There was glare, all kinds of distortion, noise etc, that while of course possible to overcome, was way out of my league to deal with. I guess one could have thrown machine learning in the mix too... :-P

ethbro 8y ago

In answer to #1, I'd assume because Chinese PMs are much the same as PMs elsewhere. PM: "Why aren't we done yet?" Dev: "Umm, we need more photos." {Cue never ending stream of request for photos}

sonnyblarney 8y ago

Physical things are hard, there are a million little issues. Sometimes 'seeing' a piece or component can reveal a lot as it represents the solution to a problem that is not otherwise evident. The smallest details like the size/type of motors, power ratings, how they are positioned, what kinds of sensors they use can reveal a lot. To boot, you're not dealing with a toy that can fall apart after 10 uses - this is gear that ostensibly has to be fairly robust, which introduces a whole other set of quality challenges beyond 'just making it work'. Example: making a little 'toy car' that ran across the floor, we ran into all sorts of control issues with the motors as the car encountered surfaces of different types, and also issues with getting caught on carpet etc.. Upon seeing how our partners made a previous device, it became pretty clear how they solved the problem. Instead of testing various different configurations and trying to find the 'best fit' - we just used their example.