Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

When a driverless car can't handle the situation

This site may earn commission on affiliate links.

woof

Fluffy Member
Supporting Member
Apr 30, 2009
1,638
2,270
So I thought I had a great idea for how to solve the problem when a driverless car get's into a sticky situation and needs help. If there's a person in the car, that's easy. But what happens when summoning a car from across the country and it get's pulled over for having a blown tail light? Or it gets into a situation it just can't handle?

Here's the idea:
Have a trained set of remote "drivers" ready to take over remotely, watching via the cameras, able to hear and speak to anyone outside the car via the audio systems. When the car signals it's stuck, it starts streaming the video feed back home, and someone will analyze the situation and "drive" the car via remote control. I'd imagine the "drivers" might have augmented reality visuals so they feel like they are sitting in the car looking out, can swivel their heads and see the view from the cameras, corrected from wide angles to human scale.

I hadn't heard anyone mention anything like this, so I hit the 'net looking to see if anyone else thought of this. Turns out they did...but not for driverless cars. It was for autonomous delivery robots roaming the sidewalks:

“We don’t want 100 percent autonomous driving. We always want 99 percent because we know and want an element of human oversight and operation at all times. We are estimating that, in the future, it will be one human operating 100 robots. There would be a bank of operators. Think of a call centre of human operators being pinged by robots in various neighbourhoods around the world, saying: ‘Ping, there’s a hole in the road, what do I do?’ or ‘Ping, there’s a new obstacle in my way, what do I do?’ A human can make that decision in a split second. There’s even a microphone and speaker on board, so nervous pedestrians can be reassured or perhaps even asked for assistance.”

Pizza, the unsung agent of the robot revolution

 
  • Like
Reactions: 1208
Being pulled over for a light? Be serious. You need to address the most serious problems first. The situations which call for immediate "rescue" need to be addressed first and quickly. When the car becomes a hazard, someone should be able to drive it to a better location. And not "remotely".
 
I think there will be very few situations it won't be able to handle... as long as a human has driven in that situation a few times, it'll be ok.
Of course there will be few situations it can't handle when compared to those it can. But for the ones it can't, it's potentially a huge problem right? You don't really need to even speculate about there being situations it can't handle, as those will be the ones requiring much more intelligence such as would be required with human interaction to assess intention or to understand instructions, rerouting strategy in an unidentifiable environment, etc.
 
I assume Tesla Network will have to have something like the OP, monitoring the driverless cars and intervening if a car calls for help. They could arrange for a tow to a SvC center, or perhaps guide the car remotely. Not to mention insuring the driverless car. Lots of stuff to think about.
 
Umm... How do you propose someone driving a driverless car to a better location, if not remotely?

You could take control remotely, but full sense of suroundings and feel for car behavior will be way too limited without a lot of additional gear installed. Prob need some VR setup with an external mic. Can't rely on that to work reliably over today's wireless networks. Also can't wait for a tow truck if the driverless car is causing traffic conditions that prohibit tow truck access.
 
You could take control remotely, but full sense of suroundings and feel for car behavior will be way too limited without a lot of additional gear installed. Prob need some VR setup with an external mic. Can't rely on that to work reliably over today's wireless networks. Also can't wait for a tow truck if the driverless car is causing traffic conditions that prohibit tow truck access.
Driverless cars will probably try to find the edge of the road to pull off if they can't handle something... It'd be extremely rare.
 
Driverless cars will probably try to find the edge of the road to pull off if they can't handle something... It'd be extremely rare.
of course the car would pull over if there was such a place to do so. However, in many cases the problem is recognizing that a road is under human traffic direction. This happens all the damn time for various reasons - events, weather, malfunctioning light, a broken down vehicle, construction where you're directed across double lanes into a single lane shared by two-way traffic,
road work, police stop, etc.
 
of course the car would pull over if there was such a place to do so. However, in many cases the problem is recognizing that a road is under human traffic direction. This happens all the damn time for various reasons - events, weather, malfunctioning light, a broken down vehicle, construction where you're directed across double lanes into a single lane shared by two-way traffic,
road work, police stop, etc.
Agreed. It needs a backup and it cant be remotely because there could be MANY vehicle that need help at the same time and they need it NOW. Not in 2 minutes, not even in 2 seconds. That's why I would be perfectly fine with rising in a self driving car but not in one that didn't have human controls still in it so I could take control when I want/need to.
 
- Driving past peopel on a horses
- Driving onto a ferry with human signals
- Driving past road works
- Driving down one lane roads where a human would have to look far away in the distance to check if any cars are oncoming
- Hazardous weather conditions which affect traction (snow, ice, heavy rain, floods..)

If anyone thinks these hurdles are easily overcome, and know the solution, please get in touch with nVidia, I'm positive they'd offer a position.
 
- Driving past peopel on a horses
- Driving onto a ferry with human signals
- Driving past road works
- Driving down one lane roads where a human would have to look far away in the distance to check if any cars are oncoming
- Hazardous weather conditions which affect traction (snow, ice, heavy rain, floods..)

If anyone thinks these hurdles are easily overcome, and know the solution, please get in touch with nVidia, I'm positive they'd offer a position.

Tesla's platform simply runs on Nvidia hardware. They've tried to make it clear that they aren't using anyone else's software package so contacting nvidia would be pointless as far as Tesla goes.

Second, own Nvidia's Davenet network running on nvidia hardware in BB8 after only 3000 miles of watching a human driver was able to drive in the rain, on a country road without markings, on highways, on dirt roads, etc. You can read the paper here: [1604.07316] End to End Learning for Self-Driving Cars

Comparatively, Tesla will have billions of miles to work with before the software is even enabled for public use.

A few questions you can ask yourself is:

  • "Has any human Tesla driver ever driven a Tesla equipped with autopilot in any one of these situations?"
  • "How many additional Tesla drivers with the new AP2 hardware will also encounter these same situations before public rollout?"
  • "Did these humans all crash and die or did they provide valuable training information in how future neural network should behave in these situations?

The ferry one does sound like a difficult task, but I'd argue that's not a driving task that has to be solved with level 5 since you're transporting the vehicle instead of the vehicle actually driving. I realize there are some places only accessible by ferry so I'm sure there will be a solution in the future but I wouldn't expect it in an initial rollout.
 
Last edited:
It is interesting to how watch detractors work. They repeatedly move the goalposts. First it was AP can't change lanes. OK, it can do that, but it can't read street signs and traffic lights, or handle city traffic ....

Now we are at the hypothetical hard case stage. Yes, some sotuations will be difficult. We can all make up difficult scenerios. We can make up impossible scenerios for human drivers, too.

The point is that AP will improve. At some point it will be good enough, not perfect but good enough. It will then continue to improve.
 
  • Like
Reactions: GSP
For me personally, every day I drive to work and I drive back home. There have been very few events out of the ordinary nearly 100% were caused by human drivers negatively impacting my planned driving path (having to drive around an accident, road closed due to accident, etc).
If I did an actual calculation this has happened about 5 times in the last 3 years so that's 5 / (365*3) = 0.4% of my actual days*.

I'd be both grateful and impressed if it simply handled 99.6% of my driving out of the box. Even though Tesla has promised level 5 capable hardware, there's nothing stopping them from rolling out level 4 software in the meantime. Although one might argue that this has the potential to confuse people when people aren't clear of the capabilities when it's not yet fully self driving 100% of the time.

*disclaimer: (yes I don't drive 365 days a year and likewise a day when I encounter a road obstruction, have many hours of normal driving as well. My calculation is just a gross estimate.)
 
Tesla's platform simply runs on Nvi...lout.

My point was, some areas are going to be very difficult where:
1. Good traction of road isn't possible, due to ice, snow, floods
2. Human to human communication is necessary.

I'm an optimist, but can see a lot of issues in the UK where we have those sorts of issues daily:
- Single lane roads where there is often communication between drivers with hand gestures or flashing light to communicate who goes first
- Horses/cyclists/walkers on the road, having to overtake them, sometimes on narrow roads with blind bends
- Floods/Snow/Ice
- Temporary road works where lane markers have to be ignored
 
I think what we'll end up seeing is zones being created where autonomy is permitted. These would be tied in with GPS coordinates, so for instance, most motorways in the UK could be manually classified as autonomous.

Over time, more sections of road, or types of areas are enabled. This would make the most logical sense to me. Start with nice and easy, get using it a ton, roll it out, find the issues, and only after then once it is very reliable, begun to enable more diverse areas like urban areas.

Where I live specifically in the UK, there are a lot of villages, really difficult extremely narrow roads which I think will be the most difficult, along with busy towns/cities.
 
  • Like
Reactions: njxman
I'm wondering when the first hit-and-run happens with a driverless car or truck. As in no humans in the car, it's completely autonomous hauling stuff perhaps, and it encounters a rare sudden edge case that it's not prepared to handle, causing it to hit a pedestrian or cyclist or or animal or other vehicle. Maybe a rock falls from a roadside cliff, damages car, car then does something bad and someone gets hurt.

Within the next 2 years? 5 years? It seems inevitable -- it will happen. Will the car know enough to stop and phone the mothership for help? Does it keep driving?

I hope it's not a Tesla.

I wonder who would be responsible? The owner of the vehicle? The maker of the vehicle?