Twitter thread by Amir Efrati at The Information. More news on Waymo. Safety drivers are back in every vehicle.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Waymo’s “commercial” ride-hailing service is... not yet what I hoped
- Thread starter strangecosmos
- Start date
lunitiks
Cool James & Black Teacher
Kind of funny that they’re adding safety drivers back to all vehicles after a safety driver hit someone on a motorcycle. Wouldn’t you want to remove more safety drivers? 
CarlK
Active Member
Kind of funny that they’re adding safety drivers back to all vehicles after a safety driver hit someone on a motorcycle. Wouldn’t you want to remove more safety drivers?
Are you trying to misunderstand Waymo on purpose?
The articles make it clear Waymo is adding secondary safety drivers and driver monitoring as the biggest thing. Given the Uber scenario it is clear how that is a responsible move for testing to add more eyes in the car, no?
For years i have made the argument that Waymo has a way less (pun intended) advanced system but a way more validated and mature system.
For example there are only 3 companies with atleast one mature subsystem and that's Waymo, Mobileye and Cruise.
Waymo has a mature hardware, sensors, integrated car platform, sensing software and planning software.
Mobileye has a mature hardware, sensors, integrated car platform, sensing software.
Cruise has a mature hardware and integrated car platform.
No one else in the industry has one or more.
Waymo having all its components in a mature stage STILL can't get to L4 show you its deficiencies that it has. Mainly its path planning.
Like Amnon said, Its not something that they can solve in the next year. Because they have been at this for 10 years, 6 years with full access to deep learning techniques.
The problem being their safety bubble as TI outline.
Amnon was BULLSEYE! when he said this was exactly the problem.
You can't drive assertively if you have a safety bubble implemented like this. You end up with a brake-happy swerving and indecisive AI like below. Leading to hundreds of rear ends and so on plus inability to function in regular flow of traffic.
This is where i see Mobileye catching up quickly in path planning and eclipsing Waymo.
Failures like these are an indictment of Waymo's path planning software. its way too brittle.
Failing to merge #1
Nitin Gupta on Twitter
Failing to merge #2
While aptiv isn't a forerunner in SDC. This ride kinda showcases how smooth their system is. Waymo for example won't be able to handle these traffic conditions. Notice how the car drives naturally as a human even in close proximity traffic. Even when it senses danger in-front of it, it never hard brakes but handle things gracefully. It operates without the aggressive safety bubble that waymo cars seem to have.
Yes i know its wayyyyyy less mature/validated compared to Waymo but it showcases the difference in path planning.
For example there are only 3 companies with atleast one mature subsystem and that's Waymo, Mobileye and Cruise.
Waymo has a mature hardware, sensors, integrated car platform, sensing software and planning software.
Mobileye has a mature hardware, sensors, integrated car platform, sensing software.
Cruise has a mature hardware and integrated car platform.
No one else in the industry has one or more.
Waymo having all its components in a mature stage STILL can't get to L4 show you its deficiencies that it has. Mainly its path planning.
Like Amnon said, Its not something that they can solve in the next year. Because they have been at this for 10 years, 6 years with full access to deep learning techniques.
The problem being their safety bubble as TI outline.
Amnon was BULLSEYE! when he said this was exactly the problem.
You can't drive assertively if you have a safety bubble implemented like this. You end up with a brake-happy swerving and indecisive AI like below. Leading to hundreds of rear ends and so on plus inability to function in regular flow of traffic.
This is where i see Mobileye catching up quickly in path planning and eclipsing Waymo.
Failures like these are an indictment of Waymo's path planning software. its way too brittle.
Failing to merge #1
Nitin Gupta on Twitter
Failing to merge #2
While aptiv isn't a forerunner in SDC. This ride kinda showcases how smooth their system is. Waymo for example won't be able to handle these traffic conditions. Notice how the car drives naturally as a human even in close proximity traffic. Even when it senses danger in-front of it, it never hard brakes but handle things gracefully. It operates without the aggressive safety bubble that waymo cars seem to have.
Yes i know its wayyyyyy less mature/validated compared to Waymo but it showcases the difference in path planning.
Last edited:
@Bladerskb developing a self driving car in California and Phoenix suburbia can definitely be different from downtown Jerusalem.
Cool to see if MobilEye can surpass Waymo with their taxi progress. It will be an interesting comparison.
Cool to see if MobilEye can surpass Waymo with their taxi progress. It will be an interesting comparison.
What I mean is NN training and data gathering for NN training is separate from consumer cars with all the major players Tesla included at least to the extent I described. Which is why Tesla in my view has no volume advantage in NN training over anyone else.
Teaching the cars to see and to drive — to the extent NNs do that — is done with engineering car data in my view, not with consumer car data. Consumer cars including Teslas do not ”learn” and they do not feed the NN directly is my view. They get static NN files through software updates that are trained at Tesla using engineering car data and simulations. They send back limited data that is not comprehensive enough to use for NN training as is.
You disagree, that is fine. Just explaing my position.
Validation or things like spotting labelling errors can of course happen with data collected from consumer cars and things like HD mapping definitely as they do not require similar data volumes as NN training. Tesla has a deployment and testing advantage with their fleet — not NN training advantage in my view.
@Bladerskb developing a self driving car in California and Phoenix suburbia can definitely be different from downtown Jerusalem.
Cool to see if MobilEye can surpass Waymo with their taxi progress. It will be an interesting comparison.
Interesting thing is that Amnon will ship the same cars from Israel to Las Vegas for CES. If Mobileye can without making any changes just drops the car on the street, hit the button and it works? then its game over!
Latest info from The Information (excerpt):
"Waymo has only weeks to meet its self-imposed deadline to launch a public taxi service using fully automated cars by the end of 2018. And right now, that deadline looks tough for the company to meet.
The Information has learned that within the past month or so, due to concerns about safety, the Alphabet company put so-called “safety drivers” back behind the wheel of its most advanced prototypes, ending a year-long period in which those people generally sat in the passenger or back seat.
Meanwhile, The Information also has learned that Waymo is only testing its most advanced prototypes in about 60 square miles, or roughly 5% of the Phoenix metropolitan area, say people with knowledge of the situation. These cars have Waymo’s most stable and vetted software, and which are supposed to be the basis for the fully automated robotaxis. That previously unreported figure suggests that any service launched this year will be severely limited for some time."
"Waymo has only weeks to meet its self-imposed deadline to launch a public taxi service using fully automated cars by the end of 2018. And right now, that deadline looks tough for the company to meet.
The Information has learned that within the past month or so, due to concerns about safety, the Alphabet company put so-called “safety drivers” back behind the wheel of its most advanced prototypes, ending a year-long period in which those people generally sat in the passenger or back seat.
Meanwhile, The Information also has learned that Waymo is only testing its most advanced prototypes in about 60 square miles, or roughly 5% of the Phoenix metropolitan area, say people with knowledge of the situation. These cars have Waymo’s most stable and vetted software, and which are supposed to be the basis for the fully automated robotaxis. That previously unreported figure suggests that any service launched this year will be severely limited for some time."
Waymo starts the paid taxi service on time:
I rode in Waymo’s new self-driving taxi service
Safety drivers are there but passive onlookers unless needed. Looks pretty good to me. Can’t see anyone else in the world doing better at this time.
I rode in Waymo’s new self-driving taxi service
Safety drivers are there but passive onlookers unless needed. Looks pretty good to me. Can’t see anyone else in the world doing better at this time.
CarlK
Active Member
Like I always say demo's and presentations do not count. Everyone could do those as impressive as they could. You need to put products in customers' hands for them to use and examine. Let's see if Waymo's latest endeavor can be qualified as a real product. Safety drivers aside it needs to take any customer who wants to ride on it to anywhere Waymo says the car can go and without needing to sign a NDA.
Last edited:
Paid program as far as I know does not require an NDA. Certainly the numbers are very small but it does seem to be doing a legit thing.
There is a vast difference between a pilot program and a demo too. This one seems to be shaping up to be a pilot program. One could argue the Early Riders was a pilot program already but of course it had the NDA.
Similar threads
- Article
