4 ms·

Everybody knows Uber is a shitshow. I'm not sure what it is that turned you off Zoox. Sebastien Thrun, who started the Google self-driving car project considere

Everybody knows Uber is a shitshow. I'm not sure what it is that turned you off Zoox. Sebastien Thrun, who started the Google self-driving car project considered Jesse Levinson to be his brightest student, and Jesse is now CEO of Zoox. The thing about Waymo is that they haven't made any real progress in about 3 years. I mean, they've rolled out hundreds of new vehicles in several cities, they've racked up 10 million real world test miles and 8 billion simulated test miles, and have discovered thousands of new edge cases, but they still can't reliably navigate certain everyday driving scenarios such as left turns at uncontrolled intersections. Behavior modelling isn't sufficient to read deeply enough into the matrix of dynamic, open world driving environments for self driving cars to perform gracefully beyond a certain threshold of xomplexity. The ML knowledge of how to deal with these kinds of problems doesn't exist, that is, unless somebody has an ace up their sleeve. If anybody has an ace up their sleeve, it is likely to be either Aurora or Zoox. Though it's more likely that in a few years time we'll have a half dozen companies all bumping up against the same glass ceiling Waymo has been trying to break through for years now.

ghaff 8y ago
>bumping up against the same glass ceiling Waymo has been trying to break through for years now I'd be happy to be proven wrong but I strongly suspect that over the next couple of years it's going to become apparent that tests and proof of concepts in a combination of limited and easy conditions (e.g. Chandler AZ) and other environments with a human backup are a very long way from scaling and generalizing to selling a truly self-driving car to the public. It's also possible we'll see assistive driving systems for certain conditions (e.g. selected freeways) that don't require human attention and that full autonomy will be relegated to a maybe someday future. From the perspective of the car manufacturers, full highway autonomy actually seems like a big win even if it's not as sexy as autopilot under all conditions.
- visarga 8y ago
  If they could solve two scenarios I would be content: highways and congested city traffic (where cars mostly sit and advance a few meters at a time). Especially the congested city traffic scenario would spare me of a ton of stress. If I don't have to watch the roads I can detach emotionally and find something else to kill time.
  - ghaff 8y ago
    Unfortunately, congested city driving is probably one of the toughest cases. Once you can drive through Manhattan or Boston you can probably handle just about anyplace with paved roads in the US [ADDED: assuming a sufficient degree of mapping and testing in those other areas].
    - true_tuna 8y ago
      Cruise has been operating in SF for a while. They seem to be doing quite well. I’ve never seen them make an unsafe or questionable move. I even did a live safety test of one, jumped off the curb at a speed and direction that would have taken me in front of the car if I’d continued. It slowed significantly, so it could stop if I had continued in front (which I did not). The safety driver was like “wtf car, what are you doing?” till he saw me and then he was kinda pissed. The Uber cars were a shit show. They didn’t know how to yield to pedestrians in the crosswalk. I’m glad the ca dmv revoked their registrations.

ghaff 8y ago

>bumping up against the same glass ceiling Waymo has been trying to break through for years now I'd be happy to be proven wrong but I strongly suspect that over the next couple of years it's going to become apparent that tests and proof of concepts in a combination of limited and easy conditions (e.g. Chandler AZ) and other environments with a human backup are a very long way from scaling and generalizing to selling a truly self-driving car to the public. It's also possible we'll see assistive driving systems for certain conditions (e.g. selected freeways) that don't require human attention and that full autonomy will be relegated to a maybe someday future. From the perspective of the car manufacturers, full highway autonomy actually seems like a big win even if it's not as sexy as autopilot under all conditions.

visarga 8y ago
If they could solve two scenarios I would be content: highways and congested city traffic (where cars mostly sit and advance a few meters at a time). Especially the congested city traffic scenario would spare me of a ton of stress. If I don't have to watch the roads I can detach emotionally and find something else to kill time.
- ghaff 8y ago
  Unfortunately, congested city driving is probably one of the toughest cases. Once you can drive through Manhattan or Boston you can probably handle just about anyplace with paved roads in the US [ADDED: assuming a sufficient degree of mapping and testing in those other areas].
  - true_tuna 8y ago
    Cruise has been operating in SF for a while. They seem to be doing quite well. I’ve never seen them make an unsafe or questionable move. I even did a live safety test of one, jumped off the curb at a speed and direction that would have taken me in front of the car if I’d continued. It slowed significantly, so it could stop if I had continued in front (which I did not). The safety driver was like “wtf car, what are you doing?” till he saw me and then he was kinda pissed. The Uber cars were a shit show. They didn’t know how to yield to pedestrians in the crosswalk. I’m glad the ca dmv revoked their registrations.