Autonomous Car Progress

[diplomat33]

Thanks for the video. I'm happy for Nvidia. But I can't get too excited for them, because my production car already does all that.

Yeah, it does appear from that video that Nvidia just did their own version of Tesla's Navigate on Autopilot. They purposely don't engage autonomous mode until they get on the highway and basically just show off lane keeping, auto lane changes and taking exits, on the highway, just like NOA.

 

[mongo]

TheInformation has another article about WayMo. But its behind a paywall.

Here is the twitter thread ...

Amir Efrati on Twitter

In Silicon Valley, where Waymo's employees take rides as if they were customers, it's worse. Riders cited problems in ~half of rides, e.g. one said they wanted to "barf." One car recently got stuck in Waymo's HQ parking lot b/c it thought a

was a

.​

Ah yes, the Pinocchio edge case.

 

[diplomat33]

TheInformation has another article about WayMo. But its behind a paywall.

Here is the twitter thread ...

Amir Efrati on Twitter

In Silicon Valley, where Waymo's employees take rides as if they were customers, it's worse. Riders cited problems in ~half of rides, e.g. one said they wanted to "barf." One car recently got stuck in Waymo's HQ parking lot b/c it thought a

was a

.​

The article is free, you just have put in an email address.

But I have copied and pasted the general findings:

Here are key findings from the Waymo data:

• Smoothness. In about 46% of Waymo’s Phoenix trips and 39% of its Silicon Valley–based trips analyzed by The Information, passengers described their rides as “smooth” in their feedback to Waymo. “It is much smoother now than a few months ago. Much smoother transition, not as jerky / rough as before,” wrote one Phoenix rider.
• Brakes, Lurches, Lane Changes, Cautious Movements. Many rides weren’t smooth though, according to passengers. In 10% of the Phoenix rides and 25% of the Silicon Valley rides, passengers reported feeling “uncomfortable” or complained about “jerky” steering and “hard braking,” “cautious” driving by the vehicle, or “accelerating too quickly.” Said a Waymo rider in Mountain View, who is also an employee: “Excessive braking for almost the entire ride. I got very carsick. Almost pressed the pullover button so I could barf.”

One recent Phoenix rider said their Waymo van “proceeded to do an emergency brake style stop and send all of us slamming into our seatbelts. It caused quite a bit of pain from being whiplashed forwards unexpectedly. I am pregnant and am concerned with how hard it was slamming on the brakes for no reason at all. My husband is also an early rider and said that this was the worst ride he'd taken.”

• Pickups & drop-offs. In about 7% of all rides, the vehicles either didn’t pick up or drop off passengers in the requested locations—or both. In some cases, the vehicles were off by 50 feet, and in at least one case, a van dropped off a rider in Tempe, Ariz., a half-mile from where they expected to be let off. “Dangerous pick up,” wrote one Phoenix rider. “The car moved before we have our one year old daughter fully buckled in. We had not yet pressed the start ride button.”
  Another Phoenix-area rider, who said she was “vision impaired,” said she was picked up by a van from the grocery store with 40 pounds of groceries. But she was dropped off nearly a quarter-mile from her home—the intended destination. She wrote: “It’s not even on the same...road. Right now it is 97 degrees and 45 percent humidity and it took me almost 15 minutes with FORTY POUNDS OF GROCERIES to get home. I had to stop 3 times to rest. It’s been about 15 minutes since I got home and I’m STILL [pouring] sweat. You HAVE to let your drivers override the car in situations like this. I’m 53 years old and this is DANGEROUS.”
• Stuck. In about 2.5% of Phoenix rides and 6.5% of Silicon Valley rides, Waymo vehicles stood still for a long period of time before either the human driver took over or a Waymo representative monitoring the vehicle from a remote location helped the car figure out how to start moving again. One Waymo rider who preferred to remain anonymous told The Information that during three trips in one week this summer, the Waymo vehicle got stuck each time. Another rider, Kelly Carrizosa, a Phoenix-area school district administrator, told The Information she sat for 15 minutes in a stopped Waymo van that was paralyzed, she said, because it seemed to believe a tree branch was a car. The human driver eventually took over.
  
• Manual driving. Passengers in Phoenix said that in more than 2% of the rides, the autonomous driving system wasn’t operating for some or all of their trip, so human backup drivers took over.
  
• Cyclists & pedestrians. In about 20 Phoenix trips and 50 in Silicon Valley, representing less than 0.5% of trips and 1% of trips in those areas, respectively, passengers reported hard braking—which they deemed unnecessary—in the vicinity of cyclists or pedestrians. Several times, Waymo passengers in Silicon Valley said there was uncomfortable acceleration toward a cyclist or reported a “close call,” implying a near collision. Other times, passengers in Silicon Valley and Phoenix said the Waymo vehicles blocked bike lanes. “Our waymo had to swerve hard to avoid a bike rider in the bike lane. Surprised the sensors didn’t pick this up,” said one of Waymo’s Phoenix riders, implying that the human backup driver in the Waymo van intervened to prevent a potential collision.
  Pedestrians also remain a trouble spot for the Waymo vehicles. “Almost hit a [couple] pedestrians and cars hesitated when it didn’t need to,” said a Phoenix customer.
  
• Long routes. Passengers reported problems with the routes taken by Waymo vans in nearly 7% of trips in Phoenix and 5% of trips in Silicon Valley. A typical complaint, from a Waymo employee passenger in Silicon Valley: “Rerouted several times along the way, adding 20 minutes to an 18 minute drive.” One factor that can affect routes is that Waymo’s vehicles are programmed to avoid potentially problematic intersections and turns depending on the time of day. At other times, vehicles can miss turns because they are unable to get to a turn lane.
  
• Rain & Wind. Several passengers noted that the vehicle didn’t seem to be able to drive autonomously while it was raining outside. Even light rain seemed to cause problems, as did puddles that collected on roads after rain. Said a Phoenix rider about their van: “Couldn't handle the water on the side of the road as we drove in the right lane...and it suddenly turned right and almost ran itself onto the sidewalk and then the [human backup] driver had to take over.” Last week Waymo said it was testing vehicles without passengers in Florida, specifically to improve at handling rain.
  
• Parking lots. In nearly 1.5% of Phoenix trips, riders complained about Waymo’s handling of a parking lot, which are a key feature of driving in suburban Phoenix. They pose challenges for Waymo vehicles because of the tight turns and presence of pedestrians. They are also a problem in Silicon Valley—including in Waymo’s literal backyard. “The car got stuck next to a plant in the RLS1 [Waymo headquarters] parking lot that it thought was a pedestrian,” said one rider.

 

[EVNow]

The article is free, you just have put in an email address.

IIRC, only a few articles are free.

 

[EVNow]

Elon Musk’s Physics Mindset May Underpin A Misguided Timetable For Full Self-Driving Teslas

One more article on Musk & FSD. This time the author speculates Musk's "physics Mindset" is bad.

 

[EVNow]

passengers reported hard braking—which they deemed unnecessary—in the vicinity of cyclists or pedestrians

Pedestrians also remain a trouble spot for the Waymo vehicles. “Almost hit a [couple] pedestrians and cars hesitated when it didn’t need to,” said a Phoenix customer.

So, Waymo has similar phantom braking issue - and also may try to hit pedestrians.

I thought Lidar solved those issues

Another rider, Kelly Carrizosa, a Phoenix-area school district administrator, told The Information she sat for 15 minutes in a stopped Waymo van that was paralyzed, she said, because it seemed to believe a tree branch was a car. The human driver eventually took over.

Well, if the tree was to the side - not sure why Waymo would stop. Also, how will Waymo handle statues if it thinks tree is a pedestrian about to step onto the road ?​

 

[diplomat33]

So, Waymo has similar phantom braking issue - and also may try to hit pedestrians.

I thought Lidar solved those issues

I guess not.

The truth is that the type of sensor probably doesn't really matter all that much. Of course, different sensors have different pros and cons but both LIDAR and cameras collect important and useful data about the objects around the car. Ultimately, what really matters is what the car does with that information.

 

[scottf200]

Yeah, it does appear from that video that Nvidia just did their own version of Tesla's Navigate on Autopilot. They purposely don't engage autonomous mode until they get on the highway and basically just show off lane keeping, auto lane changes and taking exits, on the highway, just like NOA.

Big big difference tho is that they are using/requiring HD maps. (Stated here at 1m49sec / 109 seconds in: ...YT /1W9q5SjaJTc?t=109 )
Both Tesla and Comma.AI were going to use (rely on) HD maps many months ago but both realized that the real world is too dynamic (crashes in the way, road cleanup/sweeping, construction, etc).

 

[arghx7]

Sounds like competitor's self driving efforts have as many issues as Tesla's but they rely on expensive and cumbersome Lidar and HD maps. So I guess that means Tesla is ahead...? Similar performance with lower cost and more cars on the road that are capable (HW2.5 or HW3).

 

[Tam]

...Pickups & drop-offs. In about 7% of all rides, the vehicles either didn’t pick up or drop off passengers in the requested locations—or both...

I think it has solved this problem by telling riders to wait at its own station signs (just like traditional bus stations):

Photo: Automotive News

...she was dropped off nearly a quarter-mile from her home—the intended destination. She wrote: “It’s not even on the same...road. Right now it is 97 degrees and 45 percent humidity and it took me almost 15 minutes with FORTY POUNDS OF GROCERIES to get home. I had to stop 3 times to rest. It’s been about 15 minutes since I got home and I’m STILL [pouring] sweat. You HAVE to let your drivers override the car in situations like this. I’m 53 years old and this is DANGEROUS.”...

This is plain cruelty. There's a Waymo paid human backup safety driver there. Human should act more responsibly, more humanely than the machine does. Even if it means they get fired. The way I see it is: Earning money from work is to help us to avoid committing acts of cruelty like this.

This kind of cruelty to human riders should not be replicated in technology and should not be allowed.

 

[diplomat33]

This is plain cruelty. There's a Waymo paid human backup safety driver there. Human should act more responsibly, more humanely than the machine does. Even if it means they get fired. The way I see it is: Earning money from work is to help us to avoid committing acts of cruelty like this.

This kind of cruelty to human riders should not be replicated in technology and should be allowed.

I completely agree. And people complain about Enhanced Summon awkwardly just stopping a few feet from the waypoint on the map. At least ES does not stop 1/4 mile from your intended destination!

 

[johnm]

And people complain about Enhanced Summon awkwardly just stopping a few feet from the waypoint on the map. At least ES does not stop 1/4 mile from your intended destination!

And my paper airplane has never killed anyone unlike a 737 MAX 8. Enhanced summon cannot end a quarter mile from your destination, because it is limited to driving ?00 feet at a time. Without the restrictions, would it ever end up a quarter mile from where it was supposed to be? I would bet yes (probably not super often, but probably not never).

 

[diplomat33]

And my paper airplane has never killed anyone unlike a 737 MAX 8. Enhanced summon cannot end a quarter mile from your destination, because it is limited to driving ?00 feet at a time. Without the restrictions, would it ever end up a quarter mile from where it was supposed to be? I would bet yes (probably not super often, but probably not never).

Yes but which is worse, to have ES stop a few feet from it's waypoint or to be in an autonomous vehicles that drops you off a 1/4 mile from your destination on a hot summer day with groceries, to walk home on foot? Clearly, the latter.

 

[EVNow]

I think it has solved this problem by telling riders to wait at its own station signs (just like traditional bus stations):

Work around would be a charitable way of putting it, rather than "solve".

 

[emmz0r]

Why did the human driver take 15 minutes to take over. Was he/she totally knocked out cold or something.
These stories are hilarious dude. How fresh are they?

 

[EVNow]

Why did the human driver take 15 minutes to take over. Was he/she totally knocked out cold or something.
These stories are hilarious dude. How fresh are they?

I guess Waymo has told them to wait for x minutes before taking over.

The question about when this happened is a good one. May be the worst examples are from many months ago and have been fixed now.

BTW, @diplomat33 does the article have any reaction from Waymo ?

 

[diplomat33]

I guess Waymo has told them to wait for x minutes before taking over.

The question about when this happened is a good one. May be the worst examples are from many months ago and have been fixed now.

BTW, @diplomat33 does the article have any reaction from Waymo ?

No. The article that I read just gave the results of the anonymous feedback.

 

[EVNow]

No. The article that I read just gave the results of the anonymous feedback.

Then, it is "poor" journalism. Why not give Waymo a chance to respond.

 

[Tam]

...The question about when this happened is a good one...

It says "this summer":

"One Waymo rider who preferred to remain anonymous told The Information that during three trips in one week this summer, the Waymo vehicle got stuck each time. "

It also says:

"The passenger ratings and feedback come from rides that occurred in July and part of August—mostly in the Phoenix area—and show improvement compared to Waymo’s performance during 2,500 rides in the first quarter of the year,.."

...May be the worst examples are from many months ago and have been fixed now...

If they are not worst, they wouldn't make it to the news!

Whether they are fixed now or not, we'll have to wait for another report or Waymo could proactively publicize its solutions!

 

[acoste]

Really nice in depth video of Nvidia's autonomous vehicle progress. The video shows you exactly what the car is seeing with great commentary too:

Thanks for the vid. there are some interesting points.

At the Y crossing for one moment it thinks the gore zone is drivable and it also offers the possibility of changing lanes into the gore. Just like how Tesla accidents happen.

One good thing! As soon as the car on the right touches the divider lane (switches lane into ours), it is labeled as "in our lane". Tesla isn't very good at localizing lane changing cars in front.