Autonomous Car Progress

[emmz0r]

If the car might be safe, but you are not sure yet. So you have someone monitor until the confidence level allows unmonitored operation.

Like student drivers: passenger can't do anything but evaluate and yell. Or Tesla with nags (on a better future SW version)

Then they need to be in the car. There's no way a remote system over internet can be safe for realtime control on this level. Ping over 4g can also be terrible further increasing lagency6

 

[mongo]

Then they need to be in the car. There's no way a remote system over internet can be safe for realtime control on this level. Ping over 4g can also be terrible further increasing lagency6

What level of latency are you worried about? How long does it take a normal driver to check mirrors or tune the radio?

Having someone monitoring is less risky than not and seems like a valid step in development. For a highly untested system, they would use in vehicle person.

 

[emmz0r]

What level of latency are you worried about? How long does it take a normal driver to check mirrors or tune the radio?

Having someone monitoring is less risky than not and seems like a valid step in development. For a highly untested system, they would use in vehicle person.

Network communication over LTE fail over regular basis. What's less risky is that the safety driver is physically there. LTE is only suitable for non mission critical realtime operation.

Realtime mission critical systems needs to be local or have its own military grade communications system.

It is no better than worst case scenario and it would never be allowed to pass as a substitute for a local safety driver.

Also the video feed to the remote dude needs to be compressed locally and encoded. Then transferred. This requires several frames to be able to do compression. So only at that level it's too late. Not to mention enormous bandwidth costs

 

[SandiaGrunt]

So Tesla fans now believe that Waymo just made a remote controlled car and is just lying about achieving full autonomy.

I have a couple questions.

1) If it’s just remote control, why did they choose to do it Phoenix instead of in a harder place with more potential, like NYC? I mean, if it’s just people driving remotely, it shouldn’t matter too much where the cars are. NYC would look much more impressive and has more riders.

2) There are a bunch of little startup companies like Phantom Auto who make remote control services and try to sell them to AV companies. They made their remote control systems very quickly, maybe it under a year. Waymo took ten years to put journalists in empty cars. Is Waymo just a really slow and incompetent remote control company?

 

[CarlK]

But you have to have staff available, for the max peak journeys, otherwise you are not available to provide 1:1 monitoring of all journeys in progress.

if you are going to employ them anyway, for that "shift", it makes no difference whether they are in the car, or sat in the office waiting for a journey to start. There is a small benefit of being sat in the office in that you only need to accommodate "Peak", which is presumably somewhat less than 100% of fleet. But it is still Peak Journeys - even if that is only 90% of fleet vehicles available that's still 90% that you have to have 1:1 staff available to monitor.

I have no opinion on whether safety driver needs to be in the car, or monitoring remotely, I'm just considering that if the need is 1:1 monitoring then how many staff are needed,a nd that is one for every journey in progress, at peak.

Those extra staff can always be assigned to do other things when uses is lighter.

 

[mongo]

Network communication over LTE fail over regular basis. What's less risky is that the safety driver is physically there. LTE is only suitable for non mission critical realtime operation.

Realtime mission critical systems needs to be local or have its own military grade communications system.

It is no better than worst case scenario and it would never be allowed to pass as a substitute for a local safety driver.

Also the video feed to the remote dude needs to be compressed locally and encoded. Then transferred. This requires several frames to be able to do compression. So only at that level it's too late. Not to mention enormous bandwidth costs

Yes, physically present is better than remote. Rrmote is still better than none (assuming a non zero uptime). It would be a transition step, perhaps with fail safe on loss of communication.

Regarding latency:
2 frames at 60 fps = 33mS = 3 feet at 60 MPH.

 

[CarlK]

Network communication over LTE fail over regular basis. What's less risky is that the safety driver is physically there. LTE is only suitable for non mission critical realtime operation.

Realtime mission critical systems needs to be local or have its own military grade communications system.

It is no better than worst case scenario and it would never be allowed to pass as a substitute for a local safety driver.

Also the video feed to the remote dude needs to be compressed locally and encoded. Then transferred. This requires several frames to be able to do compression. So only at that level it's too late. Not to mention enormous bandwidth costs

When you can remotely control a drone thousands miles away this would not be too much of technical challenge. Besides we are talking about "monitored" not "controlled" although some degrees of control could be done too. As for cost I don't think that will stop Waymo from doing it. It's not a money making business for them for the foreseeable future.

 

[emmz0r]

When you can remotely control a drone thousands miles away this would not be too much of technical challenge. Besides we are talking about "monitored" not "controlled" although some degrees of control could be done too. As for cost I don't think that will stop Waymo from doing it. It's not a money making business for them for the foreseeable future.

What kinda drone are you talking about?

Yes, physically present is better than remote. Rrmote is still better than none (assuming a non zero uptime). It would be a transition step, perhaps with fail safe on loss of communication.

Regarding latency:
2 frames at 60 fps = 33mS = 3 feet at 60 MPH.

That's the encoding alone. Now add in image capture, LTE latency, server latency, local network latency at the company, congested bandwidth on LTE (yeah videos stop and lag).

Also if you have a keyframe every 2nd frame, you end up with a very low compression on the video, sending gigabytes of data . So you have a bandwidth limitation.

The point is that the latency on a normal, commercial internet and with varying degrees of bandwith and coverage, it's just not reliable. Look at Smart Summon. Everything needs to be right, and that only detects if you let go of the button. it does not send you a live video feed of 3+ cameras in full resolution.

I've worked on realtime vision systems, and even locally it's a constant battle to keep the latency low.

 

[mongo]

What kinda drone are you talking about?

Military,.They do fly the drones vs autopilot, but that example may be a bit misleading because the pilots in NV are not doing real time dogfighting/ super high precision at a ground vehicle level where 2 feet is the difference between passing safely and a head on (offset) crash. Heading or altitude changes and camera positioning can have some lag without any issue.

A Day in the Life of a US Air Force Drone Pilot - Avionics

 

[emmz0r]

Military,.They do fly the drones vs autopilot, but that example may be a bit misleading because the pilots in NV are not doing real time dogfighting/ super high precision at a ground vehicle level where 2 feet is the difference between passing safely and a head on (offset) crash. Heading or altitude changes and camera positioning can have some lag without any issue.

A Day in the Life of a US Air Force Drone Pilot - Avionics

The military drones have their own dedicated, proprietary communications system, probably sending uncompressed frames. They are not operated over LTE

 

[mongo]

The military drones have their own dedicated, proprietary communications system, probably sending uncompressed frames. They are not operated over LTE

Yup, another reason they are not both apples. However, speed of light does not change so signals going to the other side of the world still have lag. If using geostationary satellites (no idea if they do vs constellation), the lag is worse than land based links.

 

[mspohr]

Humm
Based on this Google thing Global Ping Statistics - WonderNetwork, round trip ping times between distant servers are under 0.6 second (including 2 60fps refresh). At 60 MPH that is 53 feet (ignoring human reaction time ). 80 (AZ highway) 70.4 feet.

Which all beside the point that, in the short term, they could move the safety drivers to a local operations center. Long term there would ideally not be constant monitoring per vehicle, but that was not the inital premise.

Our military does a good job of killing people with drones on the other side of the planet from a Nevada base.
I think Waymo could do it.

 

[emmz0r]

Our military does a good job of killing people with drones on the other side of the planet from a Nevada base.
I think Waymo could do it.

It's not comparable.

You're comparing military dedicated systems not needing to react on a split second to commercial systems with congested and shared links, needing to react in a split second.

How do drones overcome latency?

 

[DanCar]

Before you call him stupid why dont you google this thing called latency. Your talking about being in a totally different country not even in the same city while controlling a car going 80 mph. This is as laughable as it gets.

Just to be clear, I didn't call him stupid. I called his attack stupid.

 

[DanCar]

CarlK literally said “remotely monitoring each and every car so they could apply necessary actions when there is a need.” That’s 1:1, which would be stupid.

Remove monitoring of each and every car doesn't mean 1:1. Monitoring can be done thru software. It is so stupid to think otherwise.

 

[SandiaGrunt]

Remove monitoring of each and every car doesn't mean 1:1. Monitoring can be done thru software. It is so stupid to think otherwise.

Dude, the person I was responding to, before you got here, was @CarlK. The things he said were indeed very stupid. He really does believe that Waymo is just remote controlling cars, 1:1, despite how laughably ridiculous, pointless, and infeasible that would be.

And if the car is remotely controlled... by software... then just put the software into the car, and it’s an AV.

 

[DanCar]

Dude, the person I was responding to, before you got here, was @CarlK. The things he said were indeed very stupid. He really does believe that Waymo is just remote controlling cars, 1:1, despite how laughably ridiculous, pointless, and infeasible that would be.

And if the car is remotely controlled... by software... then just put the software into the car, and it’s an AV.

Dude, don't attack other people. When a car is driving straight on the freeway on a clear day, it can still be considered remotely monitored. Demeaning other people whether their thoughts are right or wrong is very stupid. People are laughing at you because you are likely take the same approach elsewhere and making moves that limit you. Just because you think it is infeasible doesn't mean it is. The power of the data center is much different then what you can put in a car.

 

[SandiaGrunt]

Just because you think it is infeasible doesn't mean it is.

I’m an authority on the subject.

 

[DanCar]

I’m an authority on the subject.

Cool, give us some numbers. Tells us why a car that is approaching a construction zone and isn't sure what to do, and slowing coming to a stop wouldn't be able to be remotely controlled. The remote operator would just have to give the software confidence which path to take.

 

[SandiaGrunt]

Why hasn’t anyone tried to address my questions?

1) If it’s just remote control, why did they choose to do it Phoenix instead of in a harder place with more potential, like NYC? I mean, if it’s just people driving remotely, it shouldn’t matter too much where the cars are. NYC would look much more impressive and has more riders.

2) There are a bunch of little startup companies like Phantom Auto who make remote control services and try to sell them to AV companies. They made their remote control systems very quickly, maybe it under a year. Waymo took ten years to put journalists in empty cars. Is Waymo just a really slow and incompetent remote control company?