Autonomous Car Progress

[diplomat33]

"We’re excited to take part in #AVS2020! For those tuning in, check out these events: → Today @ 8:15am PT: Waymo Safety Keynote Address ft. Tracy, Qi & Matt → Today @ 1:05pm PT: Urban Delivery Vehicles ft. Lauren → Thursday @ 10:05am PT: Delivering Freight ft. Charlie" https://twitter.com/Waymo/status/1287742014067486720

I'm thinking the screenshot has a typo and it should be July 27-30.

 

[Soda Popinski]

Thanks. And that just confirms that the current hardware is not good enough for reliable L5 because there will be cases where the cameras won't see well enough and the radar will be useless, where our cars will hit objects if the driver is not alert and paying attention. So our cars won't be able to respond appropriately to these cases on their own.

Assuming we don't expect the car to drive when a human can't, what scenarios do you envision the cameras won't be able to see a pedestrian, while a driver could? Sun blinding the camera is a scenario that pops to mind? Any others?

 

[diplomat33]

Assuming we don't expect the car to drive when a human can't, what scenarios do you envision the cameras won't be able to see a pedestrian, while a driver could? Sun blinding the camera is a scenario that pops to mind? Any others?

Another scenario might be if the cameras got obstructed by dirt or something but visibility was otherwise good. But that's not really what I am talking about. I am talking about cases where a human driver would have trouble seeing the pedestrian. My post was in response to the video I shared with the stroller in the dark tunnel and driving at night with an upcoming car blinding you with their headlights. Those are both situations where the human driver would be expected to drive in general and we would expect the autonomous car to drive as well, but where a sudden edge case makes it hard for the human to see the pedestrian in time. I think in those cases, we would want the autonomous car to be able to handle those edge cases more reliably than a human. Surely, we would not be ok with autonomous cars hitting pedestrians if a human driver would also have hit them? If we can make autonomous cars safer than humans, where they don't hit pedestrians that we might hit, that should be the goal, no? I think so.

In case you missed it, this is the video that I am referring to:

 

[daniel]

A computer does not function the way a brain does. A brain is slow but intelligent. A computer is fast but stupid. My way of approaching it is that a computer in the near future will not be able to extract as much or as reliable information from a camera image as a human brain can from vision. Added sensors provide more kinds of inputs, which the computer has the speed to process, thereby (in part) making up for its lack of intelligence.

 

[diplomat33]

• Lucid will equip its upcoming Air electric sedan with standard lidar and in-car monitoring for future Level 3 self-driving capability in what it calls DreamDrive.
• The Air will launch with less tech than what Tesla currently offers, but within a year it will likely be on par with Cadillac’s Super Cruise hands-free system and within three years, will have Level 3 hands off and eyes-off capabilities, which no automaker currently offers.
• The vehicles are not planned to become Level 4 ready because Lucid sees Level 3 as a goal it can achieve without adding additional cost to its vehicles.

Lucid Air EV Will Have Lidar and Driver-Assist Systems on Board

Lucid is powered by Mobileye so I think they have an excellent chance of pulling this off. If they do, it would mean a L3 consumer car in 3 years.

 

[diplomat33]

"About 5% of Waymo robotaxi rides are fully driverless, while the remainder operate with a safety driver, according to Schwall. "
Waymo: Human factor key to unlocking AV safety, public trust - FreightWaves

 

[Daniel in SD]

"About 5% of Waymo robotaxi rides are fully driverless, while the remainder operate with a safety driver, according to Schwall. "
Waymo: Human factor key to unlocking AV safety, public trust - FreightWaves

I wonder what the logic behind this is. Maybe they use the safety driverless vehicles only for routes that they feel are the safest?

 

[diplomat33]

I wonder what the logic behind this is. Maybe they use the safety driverless vehicles only for routes that they feel are the safest?

Yes, that is my guess. Waymo is very safety conscious. So presumably, Waymo only allows the driverless rides on the absolute safest routes. For other routes, that the Waymo car might still handle just fine, but where there is some risk, Waymo plays it safe and puts in a safety driver "just in case".

If you think about it, removing the safety driver probably has more cons than pros:

PROS
It gives you nice PR that says "look we have autonomous driving". It can be a metric for how reliable your autonomous driving is.

CONS
A driverless ride does not help you improve the system any better than a ride with a safety driver. You can get the same data to improve your system from a ride with a safety driver. And if a driverless ride ever got into an accident, there would big blow back that your system is not ready yet, maybe demands to cut back on your ride-hailing, not to mention some of the public being less willing to ride in your robotaxi.

So right now, I really don't see a huge incentive to remove the safety driver.

 

[Daniel in SD]

Yes, that is my guess. Waymo is very safety conscious. So presumably, Waymo only allows the driverless rides on the absolute safest routes. For other routes, that the Waymo car might still handle just fine, but where there is some risk, Waymo plays it safe and puts in a safety driver "just in case".

If you think about it, removing the safety driver probably has more cons than pros:

PROS
It gives you nice PR that says "look we have autonomous driving". It can be a metric for how reliable your autonomous driving is.

CONS
A driverless ride does not help you improve the system any better than a ride with a safety driver. You can get the same data to improve your system from a ride with a safety driver. And if a driverless ride ever got into an accident, there would big blow back that your system is not ready yet, maybe demands to cut back on your ride-hailing, not to mention some of the public being less willing to ride in your robotaxi.

So right now, I really don't see a huge incentive to remove the safety driver.

The main benefit I see to getting rid of the safety driver is that then you never have to run any simulations to see what would have happened if the safety driver hadn't intervened. I guess it really depends on how accurate they think their simulations are.

 

[diplomat33]

The main benefit I see to getting rid of the safety driver is that then you never have to run any simulations to see what would have happened if the safety driver hadn't intervened. I guess it really depends on how accurate they think their simulations are.

I see a few problems with this idea:
1) It only works if there would have been disengagements with a safety driver. If the ride has no disengagements with a safety driver, then removing the safety driver won't give different results.
2) It would also be potentially reckless. You can't just put an autonomous car on the road with no safety driver and just see what happens. What if the robotaxi does cause a serious accident?
3) In the case of Waymo, the rides are not just test rides. They are ride-hailing rides with passengers in the backseat. So there is an added safety concern there. If you remove the safety driver and there is an accident, you could potentially cause injury to your passengers.
4) Like you said, if the simulations are good enough, then simulations would be a better option than removing the safety driver just to see what happens.

Your method could potentially work in a a fake town, set up just for testing. But I don't think it would be safe to do this method in the real-world with other people around and even less so, with passengers in the back seat. In the real-world, you need to have a safety driver if there is any risk of injury or accident to protect other drivers and pedestrians.

 

[willow_hiller]

Are we certain that Waymo is running the exact same firmware on vehicles with and without safety drivers?

It could be that they test new builds of their firmware with safety drivers for X miles before allowing the driverless fleet to trial it.

 

[diplomat33]

Are we certain that Waymo is running the exact same firmware on vehicles with and without safety drivers?

It could be that they test new builds of their firmware with safety drivers for X miles before allowing the driverless fleet to trial it.

True. That would make sense. Similar to Tesla's Early Access with new features. Try the "beta" stuff with safety drivers and then when it is fully validated with enough miles and you know the software is good enough, then remove the safety driver.

 

[powertoold]

Waymo's not getting anywhere fast enough. My cynical guess about why 5% have no safety driver is that they still need reliable low-latency remote access to the car, and only 5% of the cars drive in these areas.

When you start to understand the real challenges with fsd (mostly involve predicting the future, not about sensing the current), much of the "progress" these companies are making is smoke and mirrors.

 

[diplomat33]

Waymo's not getting anywhere fast enough. My cynical guess about why 5% have no safety driver is that they still need reliable low-latency remote access to the car, and only 5% of the cars drive in these areas.

That is a cynical guess indeed, based on no evidence whatsoever.

And how many Tesla drives are driverless?

When you start to understand the real challenges with fsd (mostly involve predicting the future, not about sensing the current), much of the "progress" these companies are making is smoke and mirrors.

Waymo is well aware of this challenge. Waymo cars are able to anticipate all kinds of future behavior based on a variety of cues.

Here is a Waymo presentation where Anguelov talks about prediction:

 

[Daniel in SD]

2) It would also be potentially reckless. You can't just put an autonomous car on the road with no safety driver and just see what happens. What if the robotaxi does cause a serious accident?

A robotaxi with a safety driver is 100% useless. Eventually you have to remove the safety driver and see what happens! Obviously they should only do that if they've calculated that it meets their safety goal.

 

[diplomat33]

A robotaxi with a safety driver is 100% useless. Eventually you have to remove the safety driver and see what happens! Obviously they should only do that if they've calculated that it meets their safety goal.

Then, I guess we can assume that ~5% of the rides do meet Waymo's safety goals.

 

[daniel]

I want to ride in a driverless car.

 

[diplomat33]

I want to ride in a driverless car.

Me too.

 

[daniel]

• Lucid will equip its upcoming Air electric sedan with standard lidar and in-car monitoring for future Level 3 self-driving capability in what it calls DreamDrive.
• The Air will launch with less tech than what Tesla currently offers, but within a year it will likely be on par with Cadillac’s Super Cruise hands-free system and within three years, will have Level 3 hands off and eyes-off capabilities, which no automaker currently offers.
• The vehicles are not planned to become Level 4 ready because Lucid sees Level 3 as a goal it can achieve without adding additional cost to its vehicles.

Lucid Air EV Will Have Lidar and Driver-Assist Systems on Board

Lucid is powered by Mobileye so I think they have an excellent chance of pulling this off. If they do, it would mean a L3 consumer car in 3 years.

If it's truly L3 on all the roads I normally drive on I'll buy one. But I'm wary of a car advertised as L3 based on being able to do L3 on just a few roads, and then maybe refuse to do even L2 where my Model 3 can do L2.

And of course it's vaporware until I can buy one. I've been following EVs since 2004 when my Prius would occasionally go into electric mode and I installed a very early "EV switch" in it. And there have been more EV announcements than you could shake a stick at. And other than NEVs and the Zap Xebra, I think Tesla is the only start-up company that has actually brought EVs to market. None of the many other start-up EV companies has ever actually sold a car. (I think AC Propulsion sold two cars, and CommuterCars has sold a handful. But basically the start-ups have all been failures.)

Does Lucid have the capital to actually build a car factory? Tesla was founded by billionaires and built the Roadster, by hand, on the Lotus Elise rolling chassis, to establish its credibility and raise the capital to move on to the Model S. I'm not holding my breath for Lucid.

 

[DanCar]

Seriously, are you trolling? This is getting really old.

L4 means the car is doing all the driving within a limited ODD. And that is exactly what Waymo cars do. So based on the SAE levels, Waymo has L4. And last year, Waymo reported a disengagement rate around 13k miles per disengagement. That means that Waymo safety drivers spend most drives just sitting there and don't need to do a thing.

The "why don't we see more videos" and "it's all a marketing stunt" cliches are getting old. Waymo has provided plenty of evidence but you refuse to accept it. You are playing this cute game of demanding proof that you've already decided will never be good enough. And 20M autonomous miles is not a marketing stunt!

And, yes, Waymo is open to the general public in Chandler, AZ. Last time I checked, "non employees" are part of the general public, no? And non-employees can sign up, and once accepted, can use Waymo robotaxis anytime they want. In fact, thousands of people from the general public have used Waymo robotaxis.

Stop trolling. A car that needs constant supervision is L2. Show me the plenty of evidence in the last 6 months? Oh you have none, wow. And tell me when you've ridden a driverless waymo. Oh I see you can't. Neither can 99.999% of the rest of the population. And the few that can ride don't get an unoccupied vehicle, that is reserved for the marketing stunts. You are really getting old.