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

Neural Networks

This site may earn commission on affiliate links.
Really interesting paper out from Waymo today about using imitation learning to train a self-driving car to do path planning like a human driver.

Waymo blog post | Paper | Discussion on Gradient Descent (no trolling)

Waymo trained its network on ~50,000 miles of human driving. It makes me wonder what you could do with billions of miles of human driving.

Waymo suggests using imitation learning to make naturalistically behaving simulated vehicles, which can then be used for reinforcement learning. This is an exciting idea.

I also wonder if, once imitation learning and reinforcement learning in simulation has taken you to a certain point, it would then be productive to do reinforcement learning with the real world fleet. Disengagements, aborts, and crashes would be logged and uploaded, and used as punishments. The reward function might be miles between punishments.

You are once again duplicate posting stuff @Bladerskb already posted in another thread.
 
AFAIK every traffic light system in the world has seperate locations for the STOP light and the GO light.

Within each country, the position of each type of light in the signal is legally defined.

So once you have learnt which one(s) means STOP then it doesn't matter what colour it is. If the light in that position is lit, you stop.

I am curious about how the NN will capture special states. In particular, flashing stop lights that indicate "caution" at night.
 
  • Informative
Reactions: Engr
I am curious about how the NN will capture special states. In particular, flashing stop lights that indicate "caution" at night.

The NN most likely won't handle this at all. This is more likely a "Software v1" sort of thing. Not that it couldn't be done with an NN, but that's unlikely to be the most efficient way to do it.

I mean, unless you still believe they're going to do it with an end-to-end NN, including planning and the whole shebang...
 
I've known such people. It's a shitshow. Mostly just guess and look at other drivers for clues. And they're terrified of driving into a place which has horizontal lights.

LOL @ disagrees. What exactly are you disagreeing about? That I actually am friends with colorblind people? Or they really are afraid of areas with horizontal lights? Or that those places even exist?

"I've known such people" becomes "I actually am friends with" ... :rolleyes:

I grew up with a guy who only had vision in one eye, due to a benign tumour. He also had issues with perceiving some colours, and was deaf on the same side that he was blind. But he was was good driver and the list of things that terrified him on the road did not include traffic lights.
 
  • Like
Reactions: CarlK and croman
"I've known such people" becomes "I actually am friends with" ... :rolleyes:

I'm sorry, can you clarify, are you rolling your eyes at the fact that @AWDtsla described the situation in these two clearly non-conflicting ways? Are you saying he's making it up because his story is changing? I just want to be perfectly clear about what you're saying.

I grew up with a guy who only had vision in one eye, due to a benign tumour. He also had issues with perceiving some colours, and was deaf on the same side that he was blind. But he was was good driver and the list of things that terrified him on the road did not include traffic lights.

Well, as long as we're all recounting anecdotes, let me point out that my father is colorblind, and then let me point out that this is completely irrelevant to the discussion.

Here's something that is relevant: There are multiple types of colorblindness affecting different colors and different levels of severity. There are also various other visual disorders that can affect color perception. It's a complicated issue, and anecdotes don't help all that much.
 
That color blindness has anything to do with being a good or bad driver. There is a lot to disagree about in that post.

So the colorblind people are offended? LOL the insanity!

There is only one thing to disagree with about that post, that colorblindness isn't one thing. However the dominant form of colorblindness is red-green color blindness. So you lose.

"I've known such people" becomes "I actually am friends with" ... :rolleyes:

Known said person well enough to get stories about what happens to you when you join the army and you're red-green colorblind. You pretty much can't to anything. The guy couldn't look at a bottle of Sriracha and clearly tell you which part was red and which part was green.
 
Facts About Color Blindness | National Eye Institute

Traffic lights pose challenges, since they have to be read by the position of the light. Since most lights are vertical, with green on bottom and red on top, if a light is positioned horizontally, a color blind person has to do a quick mental rotation to read it.

Thank you for finally confirming my point.

lights.png
 
Thank you for finally confirming my point.

View attachment 360348
In some countries, like here in Switzerland, you can find out which one is which when the light state changes.
Here, if the light switches from green to red it is
1. green
2. yellow
3. red
and from red to green
1. red
2. red AND yellow (means "prepare to drive but wait until green")
3. green

I know, this isn't helpful if you drive up to a light that is already in steady state.
 
In some countries, like here in Switzerland, you can find out which one is which when the light state changes.
Here, if the light switches from green to red it is
1. green
2. yellow
3. red
and from red to green
1. red
2. red AND yellow (means "prepare to drive but wait until green")
3. green

I know, this isn't helpful if you drive up to a light that is already in steady state.
Do they have a system by which an approaching emergency vehicle can short circuit the sequence to give themselves a green light, and if so, does it still cycle properly to that state or does it just jump to that state skipping the transitions?
 
Do they have a system by which an approaching emergency vehicle can short circuit the sequence to give themselves a green light, and if so, does it still cycle properly to that state or does it just jump to that state skipping the transitions?
As @CK_Stuggi said, emergency vehicles don‘t alter lights. The only ones doing that are buses and trams of public transportation arround here. They are prioritized but lights do the full cycle as described.
 
No, emergency vehicles are allowed to disregard the lights, warning other traffic participants using audible and visible cues. In Europe, that means differing kinds of loud sirens and usually blue blinking lights.
Well, that's the default case here too, but many cities have systems (in my experience most often using a line of sight system, but that may vary regionally) in place that allow emergency vehicles to pre-empty the traffic timing. They still run the lights and sirens, of course, but this helps increase safety (usually) by also forcing the intersection lights to grant right of way to the emergency vehicles, so they don't have to try and move around vehicles in their path that are otherwise stopped, etc. These won't be found at all intersections in cities that have them, just bigger / busier ones.
 
  • Informative
  • Like
Reactions: croman and hiroshiy
As @CK_Stuggi said, emergency vehicles don‘t alter lights. The only ones doing that are buses and trams of public transportation arround here. They are prioritized but lights do the full cycle as described.
In Europe emergency vehicles can change the lights. They have the same transponders as buses and trams use to get priority. But they don't always use them if not needed.
And using the priority button only shortens the cycle for crossing traffic. The signals goes thru the normal cycle.
 
  • Informative
Reactions: BioSehnsucht
Ran across an interesting article on the basics of neural networks. It gives a good overview on how these things work for amateurs like myself. From the article:

“In this piece we'll dig deep into deep learning. I'll explain what neural networks are, how they're trained, and why they require so much computing power. And then I'll explain why a particular type of neural network—deep, convolutional networks—is so remarkably good at understanding images. And don't worry—there will be a lot of pictures.”

How computers got shockingly good at recognizing images