Will Full Self Driving detect this?

[bob_p]

The focus on FSD has been on items like high resolution maps and object detection - which is enough to handle typical driving situations.

What about unusual situations?

For example, when at an intersection with a police officer manually directed traffic using a combination of a whistle and hand signals. Will the FSD software be able to detect this? Will it be able to isolate the police officer in the intersection, rule out pedestrians in the area, and listen for a traffic whistle or detect the officers hand signals?

There are other circumstances when emergency vehicles or trains are nearby - and not visible - and use audible warnings (sirens) to warn of their approach, so vehicles can slow down in advance of making visual contact. Will the FSD software be able to detect this?

When operating under EAP, it's the driver's responsibility to maintain control of the vehicle. Under FSD, there may not even be anyone inside the vehicle.

Has there been any mention by Tesla about how FSD will be able to react to these situations?

Another one... What if the software makes a mistake on the speed limit or doesn't detect a stop sign or traffic signal. A police car starts pursuing the car with a siren and lights, asking the driver to pull over. Will FSD comply - pull over and come to a complete stop? Since there might not be any driver, the officer wouldn't be able to give anyone a ticket. And when the officer has taken down the vehicle information, how will FSD detect it's OK to resume travel?

It will likely be the car's owner - not Tesla that will be held responsible for not handling these situations properly.

It's difficult enough just to get normal driving working, but these unusual situations may be just as difficult to handle - and is the AP2 sensor suite sufficient (such as listening for audible warnings)?

Since Tesla claims FSD should have sufficient hardware (sensors/processors) to get regulatory approval for full self driving (when the software is ready), should we assume Tesla has anticipated these issues and has a plan to handle them with AP2.x?

 

[Derek Kessler]

Tesla has discussed nothing about how FSD will handle anything.

So... nobody outside of Tesla knows. And few inside of Tesla know.

 

[McRat]

What I foresee is the need for a central command center that can remotely override AV controls. The car sees a situation that is outside its programming. The car stops, turns on the emergency flashers, and sends a signal to the command center. A human operator takes control to 'unstick' the car, then the AV system sends a green light to command, and the command operator goes on to the next alert. Just the issue of breakdowns will require this. Flat tires, collisions, mechanical breakage, are but a few things will most likely need a human to step in.

 

[The Duke]

"For example, when at an intersection with a police officer manually directed traffic using a combination of a whistle and hand signals. Will the FSD software be able to detect this? Will it be able to isolate the police officer in the intersection, rule out pedestrians in the area, and listen for a traffic whistle or detect the officers hand signals?"
Of course this will be done. The question is how long it will take. You will see posted that we drive with two eyes so a computer can drive with two cameras. This is mostly true, but we are massively parallel compared to computers. We can get input from the steering wheel, ears, drivers and passengers to name a few. Eventually computers can do this as well but it's a LONG way off. AP sees the car in front of it, but not its driver which is what will be needed for "unusual" situations like someone signaling a left lane change then diving to the right because they are missing their off ramp. Humans see the signal left, the driver look right and know there is something strange, AP cannot do that yet.

The good news is handling "unusual" situations 100% of the time is not necessary to be safer than human drivers. Humans are not 100% at "unusual" situations.

 

[Achilles]

I have pondered a lot of these situations. When pulling out of a driveway into the street can the front camera see to the side far enough to not pull out in front of a speeding driver? I Was driving down a street where workers had placed cones down the middle of the lane and we were supposed to drive to the far right of them on the edge. How could the car handle that? Man, it's no wonder they can't release it yet. Makes you wonder how it ever could be "ready."

 

[BerTX]

I know what my car would do in this situation...

 

[tomas]

Here's the FSD problem as I see it. I've never seen how just a few cars on the road can be FSD. Seems like all or none, because they need to be able to "see" each other in the virtual world. And, I think there are a lot of other systems that need to be changed as well as the cars. Take @bob_p's example. Standard operating procedure for police in event of emergency is to go out and direct traffic by hand. And that makes sense in a world with human drivers. In a world of computerized drivers, I would think the first thing the officer would do is pull out their network connected device, and mark the intersection in the "central" GPS map as "off line", and all of the FSD cars would automagically reroute to completely avoid the area. Even those immediately approaching would stop and turn around. Same goes for utility companies, road repairs (like @BerTX above), etc. Don't rely on flaggers or cones, just take the road out of the GPS map while you are working on it. Sure, there are some example of when this would not work so well. Take CA 1 around Big Sur (if it ever gets repaired), where there is no alternate route. But you could get to a vast majority of cases.

If ALL of the cars are FSD, I think it is possible to re-imagine all of the driving and road-related scenarios and address them. It is the hybrid world that makes FSD impossible: some FSD cars, some driven cars, and road-related processes that were designed many decades ago around drivers. While I agree that sensors in every situation can be (far) better and faster than humans, I also believe there are too many scenarios, and too many intangibles to program what the car does upon receipt of the sensor info in a hybrid world. IMO, of course.

 

[BerTX]

Here's the FSD problem as I see it. I've never seen how just a few cars on the road can be FSD. Seems like all or none, because they need to be able to "see" each other in the virtual world. And, I think there are a lot of other systems that need to be changed as well as the cars. Take @bob_p's example. Standard operating procedure for police in event of emergency is to go out and direct traffic by hand. And that makes sense in a world with human drivers. In a world of computerized drivers, I would think the first thing the officer would do is pull out their network connected device, and mark the intersection in the "central" GPS map as "off line", and all of the FSD cars would automagically reroute to completely avoid the area. Even those immediately approaching would stop and turn around. Same goes for utility companies, road repairs (like @BerTX above), etc. Don't rely on flaggers or cones, just take the road out of the GPS map while you are working on it. Sure, there are some example of when this would not work so well. Take CA 1 around Big Sur (if it ever gets repaired), where there is no alternate route. But you could get to a vast majority of cases.

If ALL of the cars are FSD, I think it is possible to re-imagine all of the driving and road-related scenarios and address them. It is the hybrid world that makes FSD impossible: some FSD cars, some driven cars, and road-related processes that were designed many decades ago. While I agree that sensors in every situation can be (far) better and faster than humans, I also believe there are too many scenarios, and too many intangibles to program what the car does upon receipt of the sensor info. IMO, of course.

I have long thought that there won't be autonomous vehicles until all cars are autonomous. Meaning it will be a LONG time. Cutting out the stupid things that humans do is imperative. The "I don't know what that guy will do" becomes "I communicated with that car and know EXACTLY what it is going to do".

I applaud the move toward autonomy because it will create safer drivers. FSD on all roads won't occur in my lifetime, in my opinion. Maybe on divided highways with mandatory speed controls.

 

[brkaus]

At some point, yes. When, and how it will work until then? No one knows.

Car2Go shut down its network, disabled and locked all the cars, when there was sleet/ice potential here in Austin. That was with human drivers.

 

[The Duke]

I have long thought that there won't be autonomous vehicles until all cars are autonomous.

Think more... There is no reason the cars have to communicate other than seeing each other. That's the way things work now, we do not talk to other drivers except through turn signals, horn, etc which can be detected by a vehicle. Police can easily be given "traffic direction" sticks like airports use to guide planes. Then cars can be taught to obey them. xBox has had gesture recognition for years now.
Keep thinking, but think about solving the problems and your estimate will change.

 

[bob_p]

V2V (vehicle to vehicle) and V2I (vehicle to infrastructure) communication can certainly help - and it wouldn't be surprising for both to be required in order to get FSD certification.

However, as pointed out - V2V will have challenges when there are any vehicles that are not participating in V2V, though all of the V2V-enabled cars could work together to track the offline vehicles, which could help.

V2I could help considerably to notify cars of issues - such as traffic re-routing, closed roads/lanes, etc. But until that's available on 100% of the roads, the onboard software still needs to detect unusual circumstances.

Police routing traffic manually occurs around road construction, to override traffic signals (such as in areas of high congestion after special events), or other emergencies - and the directions being provided to individual cars to move, turn or stop is probably not going to be covered by V2I.

Plus, there is another challenge with V2V and V2I, unfortunately, we're in a world where cybersecurity is a growing concern, which could impact the reliability of the data provided by V2V and V2I, if someone has bad intent. Can V2V and V2I be implemented in vehicles and guaranteed to operate without risk? (especially as vehicles with V2V/V2I age - and may not keep up with the latest improvements).

It's understandable the focus right now is on getting the basic self driving features working - high resolution maps and recognizing typical objects (lane lines, highway exit/entry ramps, pedestrians, vehicles, street signs/lights, ...). But even when that's done, at best that's driver assistance - not full self driving, until the software can operate without a human involved. And having someone remotely drive the car will never happen - because Tesla will not want to take any responsibility for remotely operating a car.

 

[The Duke]

We all have our opinion, but I have driven with no assistance for a long time. I see no reason a computer cannot do the same. A car will need no outside assistance when properly set up and programmed, IMO.

 

[Mediocrates]

We all have our opinion, but I have driven with no assistance for a long time. I see no reason a computer cannot do the same. A car will need no outside assistance when properly set up and programmed, IMO.

This is the fundamental struggle with autonomous driving: people try to make vehicles experience the world as humans do, because vehicles must drive where humans also drive.

Remove the human requirements/limitations, then the solution set expands to more robust possibilities.

 

[WdstockWizard]

What I foresee is the need for a central command center that can remotely override AV controls. The car sees a situation that is outside its programming. The car stops, turns on the emergency flashers, and sends a signal to the command center. A human operator takes control to 'unstick' the car, then the AV system sends a green light to command, and the command operator goes on to the next alert. Just the issue of breakdowns will require this. Flat tires, collisions, mechanical breakage, are but a few things will most likely need a human to step in.

OMG!! Really?