Tam
Well-Known Member
They are posted as requested. Thanks for the suggestion:
https://teslamotorsclub.com/tmc/posts/1722413/
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Tesla Sensor Suite vs. LIDAR
- Thread starter Mark Harvey
- Start date
-
- Tags
- Autonomous Vehicles
They are posted as requested. Thanks for the suggestion:
https://teslamotorsclub.com/tmc/posts/1722413/
stopcrazypp
Well-Known Member
This appears to be a Stanford test vehicle (see the Stanford logo sticker on the back window and the license plate holder). It also doesn't have the manufacturer plate that a Tesla test vehicle would have.Financial speculating and investing are interesting but I am more interested on how things work and how to make driving safer.
On the other hand, when Tesla says LIDAR is not necessary, that doesn't mean it can't have LIDAR.
Tesla Model S prototype with new sensors spotted near Tesla’s HQ – potentially lidar or GPS
Two months a go a conventional LIDAR was spotted:
This one has the plate cropped out, so hard to tell, but it does look like a manufacturer plate.Recently, something look like newer LIDAR pucks were spotted:
Like I said in this thread: Tesla Model S with LIDAR ‘Puck’ spotted testing near Palo Alto HQ
I don't think those are LIDAR pucks on the Model S. I think they are GPS antennas. We use some very similar GPS antennas at work, which is why I'm pretty sure they are GPS antennas. Also, if they were LIDAR pucks, they would likely be attached more securely than with duct tape. For the test results to be meaningful, you need to know exactly how the pucks are positioned, and it needs to be reproducible.
My best guess is that they are using two GPS antennas to not only determine the position, but to also determine which direction the nose is pointing. With a single antenna, the GPS can't be used to determine the direction the nose is pointing if you're stationary, moving slowly, or if you for instance reverse.
Another reason why GPS antennas is my best guess, is the positioning of the objects. If they were LIDAR pucks you would expect them to be position at the corners of the roof, just like on the BMW. These are perfectly aligned down the center line of the car. If they are GPS antennas, this makes it easy to calculate the direction the nose is pointing. But if they are LIDAR pucks, side visibility will be very poor.
I don't think those are LIDAR pucks on the Model S. I think they are GPS antennas. We use some very similar GPS antennas at work, which is why I'm pretty sure they are GPS antennas. Also, if they were LIDAR pucks, they would likely be attached more securely than with duct tape. For the test results to be meaningful, you need to know exactly how the pucks are positioned, and it needs to be reproducible.
My best guess is that they are using two GPS antennas to not only determine the position, but to also determine which direction the nose is pointing. With a single antenna, the GPS can't be used to determine the direction the nose is pointing if you're stationary, moving slowly, or if you for instance reverse.
Another reason why GPS antennas is my best guess, is the positioning of the objects. If they were LIDAR pucks you would expect them to be position at the corners of the roof, just like on the BMW. These are perfectly aligned down the center line of the car. If they are GPS antennas, this makes it easy to calculate the direction the nose is pointing. But if they are LIDAR pucks, side visibility will be very poor.
Last edited:
You really need to add hearing to that as sound info becomes very important when entering road from a blind driveway.
Tam
Well-Known Member
Now that the 11AM press conference is over, Elon Musk has just reiterated Tesla's preference for RADAR over LIDAR:
Tesla Autopilot v8 detailed: Smarter radar in 1-2 weeks
Autopilot version 8, though, will do something similar – and, indeed, Musk argues better – to LIDAR only with far more affordable radar.
“We do not envisage using LIDAR,” he says, “LIDAR does not penetrate occlusions, it does not penetrate rain, dust, or snow, whereas radar does. LIDAR doesn’t bounce, you can’t look at things in front of the car in front of you.”
Tesla Autopilot v8 detailed: Smarter radar in 1-2 weeks
Autopilot version 8, though, will do something similar – and, indeed, Musk argues better – to LIDAR only with far more affordable radar.
“We do not envisage using LIDAR,” he says, “LIDAR does not penetrate occlusions, it does not penetrate rain, dust, or snow, whereas radar does. LIDAR doesn’t bounce, you can’t look at things in front of the car in front of you.”
Tam
Well-Known Member
Sounds like Elon Musk is doing what others advised him "impossible" in term of RADAR:
Musk said he had wanted to improve Autopilot's capabilities last year but was told it was impossible to do so without incurring more "false positives," such as a car braking suddenly for a harmless tin can.
Musk said he had wanted to improve Autopilot's capabilities last year but was told it was impossible to do so without incurring more "false positives," such as a car braking suddenly for a harmless tin can.
The next two paragraphs addressed that. With crude processing, it isn't possible. With the advanced processing they are implementing, they will get a more detailed 3D picture of objects plus their motion. That should allow them to classify things.
Reeler
Decade of Pure EV Driving
No matter what is done to the software, the current sensor suite cannot do passable blind spot detection. It is good that they are improving things, but where is our AP 2.0.
hacer
Active Member
You've posted a couple of times about "blind spot detection". Do you mean that you want the car to detect when you have improperly adjusted your outside mirrors creating blind spots? That depends on driver height, seat position etc, so it is a slightly difficult problem and one better solved through education. How To: Adjust Your Mirrors to Avoid Blind Spots - Feature.
Note that with mirrors adjusted this way you may not see a car in the mirror that is directly beside you and visible in your peripheral vision. If you wear glasses you might need to slightly tilt your end before merging to see a car there.
Reeler
Decade of Pure EV Driving
I do adjust my mirrors as you suggest and it works great on my Nissan LEAF. My wife will not on her Model X so enjoys large blind spots.
What I want is to have Tesla market a feature that works horribly when compared to other manufacturers that use that term. It is simply false advertising and the current sensor suite cannot be fixed with software. You can change lanes with the other maker's implementation if you like without checking your blind spot. That is not true with Tesla.
they probably put something like a timestamp in every signature so to not confuse the oldest signal with the new signal ( else you need to wait for all the bounce before you send out another signal, in this way you can send out signal faster ), so i would say it's safe to consider that every signal is unique to the car and the instant it is sent, you don't really need too much diversification with this, hell, if you have 5 radar in your car, you need to differentiate between radars!
It's AP that is poor at blindspot detection. I had a case where I was moving right on AP and the car thought it was OK. Just as the car started moving over the line, a car zoomed into the same lane from the right and came very close to rear ending me. It would have been her fault, but while I could see her coming, override AP, and accelerate out of the way, AP never saw her coming.
A wide angle camera mounted up high around the top of the hatch would have spotted her.
I think those pucks are likely GPS too. I have read that for full autonomous driving to work, better resolution GPS than currently exists needs to be achieved. They need to know which lane the car is in on the highway. It's possible Tesla is experimenting with using two GPS receivers to get better resolution.
Either with a modified version of the same network (we don't know for sure if Tesla is using the stock Mobileeye network, unless its baked in), or on the general purpose CPU that's part of mobileye. In the latter case, there are a whole host of classification algorithms that are not based on neural networks. They also don't have to do the same kind of identification: the classes could be far coarser, such as threat and non-threat. Regardless, they obviously found somewhere to do it.
I do not believe that AP can identify a specific Tesla, or even that a vehicle is a Tesla and not some other car model.
Or maybe I am not understanding your question.
I think the question is can the car be confused by radar returns sourced from other cars. I.e. The Tesla next to you pulses its radar and your car recieves the reflections. Any radar experts know how this works?
If that is the question, than my answer is "We don't know the answer to that". But I am confident that Tesla has figured that out. Public road spaces often contain radar signals, from police radar units and increasingly from other cars and not just Teslas.
Right it is clearly a solved problem. I can think of a few ways to do it, maybe all of them wrong
Similar threads
- Article
- Article
