Nissan launches hands-off ProPilot 2.0 On-Ramp to Off-Ramp (Rivals Navigate on Autopilot)

[acoste]

What happens if the road has changed since last HD mapping ? If it can identify and not hit new or changed objects, why does it need HD mapping at all.

Every car on the network (so not only from one brand) with EyeQ4 chip harvests data for the map. As soon as there is a change, it will be reported by the car which detects the change first. The data is ~10kB/km so it works with low speed mobile connection. This would be more of an issue in deserted areas with no cellphone coverage.

The car can still drive without the map though. But having the map increases the safety quite a bit since the self driving software isn't mature yet. At these early stages it is possible that the car will fall back to the driver.

 

[Daniel in SD]

What happens if the road has changed since last HD mapping ? If it can identify and not hit new or changed objects, why does it need HD mapping at all.

If Tesla Autopilot had HD mapping the fatality where the driver drove into a gore point would not have happened. I think there have been quite a few other accidents where the system steered off the road. Obviously HD mapping has to be used in addition to a system that can identify objects (like cars!). It just helps with safety.

 

[EVNow]

Every car on the network (so not only from one brand) with EyeQ4 chip harvests data for the map. As soon as there is a change, it will be reported by the car which detects the change first. The data is ~10kB/km so it works with low speed mobile connection. This would be more of an issue in deserted areas with no cellphone coverage.

The car can still drive without the map though. But having the map increases the safety quite a bit since the self driving software isn't mature yet. At these early stages it is possible that the car will fall back to the driver.

Yes, EyeQ is crowd sourcing- but how does a car figure out whether he map has changed and does it operate differently/ less securely ?

Would be really problematic in areas with constant construction work. Not just deserted areas.

 

[EVNow]

If Tesla Autopilot had HD mapping the fatality where the driver drove into a gore point would not have happened. I think there have been quite a few other accidents where the system steered off the road. Obviously HD mapping has to be used in addition to a system that can identify objects (like cars!). It just helps with safety.

EM says they tried HD maps for a while and it was a mistake.

 

[Bladerskb]

EM says they tried HD maps for a while and it was a mistake.

So Elon musk is god? Whatever he says goes? He's alpha and omega? The final decision maker? The easy conclusion was their implementation was horrific and with Elon's track record he told them to scrap it before they had chance to right the ship. Basically the whole automatic windshield wiper scenario all over again.

There is a reason that still today there is still only one company doing crowd sourced HD Mapping in production and that's because its a hard problem. Its not easy.

If we look at Elon's track record on these type of decisions. He also said driver monitoring camera is a mistake yet 3 ppl have died because of sleeping while on AP? Also there are maybe hundreds of videos of ppl sleeping while on AP. In-fact we now have regular police reports of people being arrested for sleeping while on AP.

Not only that but he refused to even add capacitor touch to steering wheel and stuck with reading the torque input from the driver, which has been poor.

Elon hasn't been making the best decisions when it comes to semi autonomous / autonomous driving implementation and tech.

 

[banned-66611]

Overtakes automatically as well, so no-confirmation lane change.

Suddenly the Leaf 62 looks more attractive.

 

[Xcelerator]

Definitely, look how detailed the map Nissan is using:

So are you saying the car is following this map to navigate not just general direction but also where in the lanes to drive, how to avoid barricades, etc? If so, that is really not even similar to autonomous driving. Around here, highways change virtually daily with new lane striping, barriers, cones, road block signs, etc put up all the time as the highway system is in a constant state of repairs and expansion. It would be impossible to develop an HD map and get it into cars bc before it could be coded, the roads frequently would have changed. Maybe that isn't what this is?

 

[FrenchJim]

FWIW, i wouldn't technically call that particular median barrier terminal as a gore point, but I wouldn't object using that term to the wider public.

The blame will probably be apportioned 3 ways, (road owner, car maker and driver) as each of these three couldv'e reasonably made decisions that would've avoided the fatality. I will say that it does demonstrate the difference between theory and practice. And that although HD maps are not seen as superior to full autonomy, that HD mapping should've avoided that particular fatality.

The general assumption is that the vehicle maintaining flow of traffic, is doing the baseline correct action, GPS and HD mapping should avoid established structures, they are insufficient for roadworks, but the baseline is the as constructed roads.

Humans are able to navigate without an HD map so theoretically a neural network should be able to do the same from a better full 360 degree vantage point. How do you determine when to use the maps to handle the situation of transient construction and road debris versus the normal. I agree with Tesla, we need a solution that doesn't handle normal but the long tail unusual. If you can handle the unusual (without maps), then you can handle the usual hence you don't need the maps other than to optimize paths.

 

[riche]

Humans are able to navigate without an HD map so theoretically a neural network should be able to do the same from a better full 360 degree vantage point. How do you determine when to use the maps to handle the situation of transient construction and road debris versus the normal. I agree with Tesla, we need a solution that doesn't handle normal but the long tail unusual. If you can handle the unusual (without maps), then you can handle the usual hence you don't need the maps other than to optimize paths.

Agreed. Solve for the edge conditions, and the happy path will take care of itself.

 

[willow_hiller]

Every car on the network (so not only from one brand) with EyeQ4 chip harvests data for the map. As soon as there is a change, it will be reported by the car which detects the change first. The data is ~10kB/km so it works with low speed mobile connection. This would be more of an issue in deserted areas with no cellphone coverage.

The car can still drive without the map though. But having the map increases the safety quite a bit since the self driving software isn't mature yet. At these early stages it is possible that the car will fall back to the driver.

I'm rather confused by this. The vehicles are not equipped with LIDAR, correct? If they're able to detect differences in the HD map and correct them without LIDAR, why is the LIDAR-created HD map necessary in the first place?

 

[acoste]

I'm rather confused by this. The vehicles are not equipped with LIDAR, correct? If they're able to detect differences in the HD map and correct them without LIDAR, why is the LIDAR-created HD map necessary in the first place?

The map is created by cameras. Karpathy showed how they can create a 3D scene out of the video footage. This is the same thing.
This is not the same system where the car is grabbing onto a lidar based map which I believe ride sharing startups are doing.

Last year Amnon mentioned that lidars are needed for redundancy only and it's not for the maps but it's about detecting objects on the road ahead.

@EVNow I don't share your concern regarding constructions and road changes. The system needs 10 cars to pass for a map update. There is nothing faster than that for a map update.

@Xcelerator As mentioned it takes very little time to update the map. And one more good thing about this system: it copies the path of the harvester cars. This means if people keep avoiding a pothole on the road, the system will copy it.

Humans are able to navigate without an HD map so theoretically a neural network should be able to do the same from a better full 360 degree vantage point.

This mapped system is added redundancy. It still needs a stand alone self driving feature. It's just more safe than that.

 

[EVNow]

So Elon musk is god?

No - because God doesn't exist.

 

[EVNow]

@EVNow I don't share your concern regarding constructions and road changes. The system needs 10 cars to pass for a map update. There is nothing faster than that for a map update.

That is not my point.

Lets say you have HD map. The car is now driving down that road. Then what ?
- To rely on the HD map the car has to first redo the map, compare to what is stored and use the stored HD Map only if it is correct ?
- If the current situation is different from the HD Map, does the car ignore the HD maps all together and drive without the maps or does it disable pro-pilot ?

So are you saying the car is following this map to navigate not just general direction but also where in the lanes to drive, how to avoid barricades, etc? If so, that is really not even similar to autonomous driving. Around here, highways change virtually daily with new lane striping, barriers, cones, road block signs, etc put up all the time as the highway system is in a constant state of repairs and expansion. It would be impossible to develop an HD map and get it into cars bc before it could be coded, the roads frequently would have changed. Maybe that isn't what this is?

Yes - on my 15 mile commute - I encounter multiple construction zones daily and they invariably change daily (otherwise whats the point of construction !). ​

 

[cusman]

So much for Tesla being 3+ years ahead.

Nissan entering the space soon, when Tesla has been at it for 3+ years already and has already driven over 1 billion miles is exactly that.

There are autonomous Bus driving around in some European & Asian cities and autonomous trucks doing everything except last-mile driving on some routes in the US for couple of years now already. These also generally use just vision / AI. Every EV startup from Rivian to Briton to you name it are all having some kind of vision / AI based autonomous driving included in their roadmaps as well.

VW ID series of vehicles will also be launching with Level 3 autonomous driving supported when they start becoming available 2021 onwards. VW expects to have regulations passed and able to do Level 5 autonomous driving by 2025. They are also using vision / AI like Tesla and Nissan.

Tesla expects to be FSD feature complete using vision / AI by end of 2019. Even if that is too optimistic, they are at least a few years ahead of VW and Nissan and their data advantage continues to grow in terms of how the AI will handle edge cases.

 

[willow_hiller]

That is not my point.

Lets say you have HD map. The car is now driving down that road. Then what ?
- To rely on the HD map the car has to first redo the map, compare to what is stored and use the stored HD Map only if it is correct ?
- If the current situation is different from the HD Map, does the car ignore the HD maps all together and drive without the maps or does it disable pro-pilot ?

Yes - on my 15 mile commute - I encounter multiple construction zones daily and they invariably change daily (otherwise whats the point of construction !). ​

It also doesn't make much sense in a real life context. Let's say some new construction cones are placed on the edge of a road. Ten ProPilot2 vehicles pass by, and dodge the construction cones using a Neural Net instead of the HD map because it's a new obstacle. Now, since 10 vehicles have passed the HD map is updated to include that as an obstacle. Then, in the afternoon the cones are removed. Do the next 10 ProPilot2 vehicles dodge a random bit of road that once had cones? Or do they rely in their Neural Net to immediately see there's no obstacle there. And at that point, the HD map becomes functionally useless since all cars are using the Neural Nets to override erroneous HD map data.

 

[Daniel in SD]

It also doesn't make much sense in a real life context. Let's say some new construction cones are placed on the edge of a road. Ten ProPilot2 vehicles pass by, and dodge the construction cones using a Neural Net instead of the HD map because it's a new obstacle. Now, since 10 vehicles have passed the HD map is updated to include that as an obstacle. Then, in the afternoon the cones are removed. Do the next 10 ProPilot2 vehicles dodge a random bit of road that once had cones? Or do they rely in their Neural Net to immediately see there's no obstacle there. And at that point, the HD map becomes functionally useless since all cars are using the Neural Nets to override erroneous HD map data.

I think the system recognizes that there has been a large change in the map and rebuilds it. I don't think it will be dodging random cones. The first time it goes through an unmapped area it will be more prone to errors. I would think the issue would be with very subtle changes to the driving environment not massive ones.
Tesla uses maps of where overpasses are to disable (or dial back?) automatic emergency braking. It's all about trying to make the system be as conservative as possible without triggering false positives (phantom braking, swerving, etc.).

 

[acoste]

It also doesn't make much sense in a real life context. Let's say some new construction cones are placed on the edge of a road. Ten ProPilot2 vehicles pass by, and dodge the construction cones using a Neural Net instead of the HD map because it's a new obstacle. Now, since 10 vehicles have passed the HD map is updated to include that as an obstacle. Then, in the afternoon the cones are removed. Do the next 10 ProPilot2 vehicles dodge a random bit of road that once had cones? Or do they rely in their Neural Net to immediately see there's no obstacle there. And at that point, the HD map becomes functionally useless since all cars are using the Neural Nets to override erroneous HD map data.

There are 3 building blocks of their system:

Sensing is first!!! I think you assume that localization is first but no.

So the sensing block creates the drivable space and will notice changes if there are any. (removed cones for example)

The drivable paths are the building blocks of the high definition map. And the landmarks are the building blocks for the localization. It is being used as a redundancy for sensing.
When sensing the road for drivable road space, the only sensor there is the camera because it's picture based and not shape based. And the redundancy for the camera is the map.


.

 

[JohnSnowNW]

If we look at Elon's track record on these type of decisions. He also said driver monitoring camera is a mistake yet 3 ppl have died because of sleeping while on AP? Also there are maybe hundreds of videos of ppl sleeping while on AP. In-fact we now have regular police reports of people being arrested for sleeping while on AP.

And camera monitoring would have prevented the latest death? The guy initiated AP 10 seconds before the accident occurred.
So, presumably, he stopped paying attention at some point within that time frame. How exactly would a camera operated system have acted differently to ensure the driver saw the truck in that amount of time?

The answer, of course, is that it wouldn't.

 

[Eno Deb]

EM says they tried HD maps for a while and it was a mistake.

I think for a useful discussion you first have to define what HD maps actually are. Some people seem to think that this is some kind of 3D map that includes every single obstacle etc. But that is not the case today. Major industry players such as HERE and MobileEye focus on lane boundaries, landmarks to aid high-precision localization, and annotations containing interesting metadata (such as markers for traffic lights and stop signs, speed limits and other restrictions). Also note that Tesla is using maps for a number of purposes as well. From what I have seen from some of the data provided by @verygreen they seem to have at least lane-width information and annotations as well.

 

[EVNow]

I think for a useful discussion you first have to define what HD maps actually are. Some people seem to think that this is some kind of 3D map that includes every single obstacle etc. But that is not the case today. Major industry players such as HERE and MobileEye focus on lane boundaries, landmarks to aid high-precision localization, and annotations containing interesting metadata (such as markers for traffic lights and stop signs, speed limits and other restrictions). Also note that Tesla is using maps for a number of purposes as well. From what I have seen from some of the data provided by @verygreen they seem to have at least lane-width information and annotations as well.

Good call.

There is an "enhanced" map that Tesla uses. It has more metadata than normal maps like Google Map. It is not a higher precision, within few cms, map. So not HD Map.

MobilEye maps are apparently used for redundancy. So, in theory, pro-pilot doesn't need HD Maps to do their normal driving - but Nissan is conservative and won't do any driving unless HD Map is available.

Therefore, for path sensing and foresight purposes, only a highly accurate map can serve as the source of redundancy.

That still leaves open the question of how does EyeQ blend mapping data with vision data and what does it do when they are not matching. What does Nissan do when they are not matching. Nissan should actually disable their pro-pilot (or degrade it) if vision and HD Map don't match, because there is no redundancy at that point.

I understand why MobilEye is doing it - their customers are a lot more conservative than Tesla.

Road Experience Management™ (REM™) - Mobileye

ps : It is actually funny when we think about who is more conservative - between Nissan and Tesla. Nissan was taking a big risk by putting batteries that were not thermally managed in Leaf. Ofcourse, they didn't catch fire but they degraded quickly causing all kinds of issues for Nissan.

Tesla (and GM) were conservative about battery thermal management - may be because the chemistry was such that there was more chance of fire unless cooled.

When it comes to other things, Nissan is obviously more conservative (esp. now that Ghosn is gone).