Google's ReCAPTCHA v2 would force us to train their self-driving cars to access websites, but it's a bit surprising to me that they never ended up updating it to collect new kinds of data. Surely the utility of the data they're collecting now has diminished. You can only have so many labeled stop sign images in clear sunny daylight before the extra data becomes no longer useful for training a good model, or actually harms its performance in nighttime/storm/etc
Quality seems to have diminished as well as there's often cases where I have to consciously avoid giving the correct answer, because I know that only 3 of the images are supposed to have a crosswalk and there is a fourth one with a slightly less visible crosswalk and just from experience it would always punish me for selecting the extra one.
Does anyone know the difference between an "external audio receiver (EAR)" and a microphone? I'm assuming it's a euphemism in case the public thinks "microphone" sounds creepy/invasive.
Interesting they more the halved the number of cameras in this generation. The 5th gen had 29 cameras, 5 LiDARs and 6 radars. This one has 13 cameras, 4 LiDARs and 6 radars. I'm sure the boxy shape of the Zeekr vehicles influences this, but it's a still a big reduction.
If Waymo somehow manages to avoid the new China import tariffs for this vehicle, they will be much better equipped to scale cost efficiently.