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

Model S Test Mule?

Discoducky

P100DL, 2021 M3, 4 CT reservations and counting
Supporting Member
Dec 25, 2011
5,444
32,001
My mountain
Curious if you have any insight how far a suite of relatively low cost visible-light/IR cameras would be able to determine ahead with enough certainty via its machine vision. Or perhaps, in the limit of the best traditional cameras (i.e. not LIDAR) how far would it be?

Also, if you are able to answer, did you work on 'active safety' for one of the big autos thats in or near production, or a more advanced prototype demo tech (like Google's autonomous cars)?

The cameras range would be limited by resolution and the field of view/optics. Currently, consumer grade IR cameras in general are limited to 720p and are very expensive. Even with an IR illuminator for night time driving and regardless of the optics chosen the visibility would be too poor and would severely limit your maximum speed to below that of normal driving patterns.

However, cars have headlights so IR cameras aren't necessary. A typical 1920x1080 camera would allow you to distinguish objects much further out and allow you a higher top speed, whereas a 4k camera would come closer to matching a human's perception. I cannot give an exact number on safe speeds because that would depend on the risk that Tesla would be willing to accept. Choosing a camera lens sets the field of view. The camera resolution determines the number of pixels per arc minute. The number of pixels per arc minute allows you to identify objects of a minimum size. Pick a narrow field of view and you can see very far ahead but can't see anything to either side of your lane. Pick a wide field of view to look for objects to the side of the vehicle and you limit how far ahead you can resolve an object.

I can't say much about the vehicles that I worked on except that it was for the defense industry and it made Google's cars look like low tech toys.

Edit: it is also probable that the vision solution would be paired with an automotive grade radar unit such as the ones that Delphi makes. The radar could be used for longer distance large obstacle detection and speed adjustment and then cameras could be used for closer range collision detection.

- - - Updated - - -

One more observation from looking at the pictures. There are what appears to be two GPS antennas on the roof (small black pucks) in opposing corners. This is called differential GPS and would be used for an autonomous vehicle to allow you to more precisely determine the vehicle's exact orientation. In addition, the big red button is leading me to believe that this is indeed an autonomous vehicle test mule.
 
in the video it does not seem to be driving by itself because the driving is WAY to erratic (i.e. speeding up then breaking.)

huh... And I thought the opposite. No human could drive that poorly given the situation. I mean humans can drive bad, but it appears occasionally to gun it at the car in front of it in a suicidal manner and them hit the breaks. Especially at 50 seconds. Even to me (an aggressive minded driver), that was ridiculous. It looks like the driver had a massive muscle spasm. Or perhaps the jerkiness was the human taking over.
 
Alternatively, it's an early AWD without proper traction control. Note that TC modulates both throttle and regen, and it works in turn with ABS. It's like driving without TC, stability control or ABS.
TC and stability control on two drivetrains has to be more difficult.
... and it's wet outside.
Good luck.

There are some really early roadster prototype videos on YouTube showing what having none of the nice features looks like when driving on dirt and wetter roads. Looks similar.
 

eepic

Member
Supporting Member
Nov 6, 2013
911
126
Canada
The cameras range would be limited by resolution and the field of view/optics. Currently, consumer grade IR cameras in general are limited to 720p and are very expensive. Even with an IR illuminator for night time driving and regardless of the optics chosen the visibility would be too poor and would severely limit your maximum speed to below that of normal driving patterns.

However, cars have headlights so IR cameras aren't necessary. A typical 1920x1080 camera would allow you to distinguish objects much further out and allow you a higher top speed, whereas a 4k camera would come closer to matching a human's perception. I cannot give an exact number on safe speeds because that would depend on the risk that Tesla would be willing to accept. Choosing a camera lens sets the field of view. The camera resolution determines the number of pixels per arc minute. The number of pixels per arc minute allows you to identify objects of a minimum size. Pick a narrow field of view and you can see very far ahead but can't see anything to either side of your lane. Pick a wide field of view to look for objects to the side of the vehicle and you limit how far ahead you can resolve an object.

I can't say much about the vehicles that I worked on except that it was for the defense industry and it made Google's cars look like low tech toys.

Edit: it is also probable that the vision solution would be paired with an automotive grade radar unit such as the ones that Delphi makes. The radar could be used for longer distance large obstacle detection and speed adjustment and then cameras could be used for closer range collision detection.

- - - Updated - - -

One more observation from looking at the pictures. There are what appears to be two GPS antennas on the roof (small black pucks) in opposing corners. This is called differential GPS and would be used for an autonomous vehicle to allow you to more precisely determine the vehicle's exact orientation. In addition, the big red button is leading me to believe that this is indeed an autonomous vehicle test mule.

cool, thanks for the insight! do you have a sense of polling or refresh rates typically used for automotive purposes to balance processing complexity/requirements vs adequate reaction time? I could imagine the order of magnitude being anywhere from 1Hz to 1kHz
 

DaveT

Searcher of green pastures
Nov 15, 2012
3,502
11,184
Texas
One more observation from looking at the pictures. There are what appears to be two GPS antennas on the roof (small black pucks) in opposing corners. This is called differential GPS and would be used for an autonomous vehicle to allow you to more precisely determine the vehicle's exact orientation. In addition, the big red button is leading me to believe that this is indeed an autonomous vehicle test mule.

+1. My vote is this is Tesla testing out cameras and sensors for autopilot.
 
cool, thanks for the insight! do you have a sense of polling or refresh rates typically used for automotive purposes to balance processing complexity/requirements vs adequate reaction time? I could imagine the order of magnitude being anywhere from 1Hz to 1kHz

Sorry, I must have missed this post before. I can't go into the specifics about the rates. Sensors are typically run very fast. Vehicle control loops are typically run much slower, though still much faster than human response time in order to be considered safe.
 

About Us

Formed in 2006, Tesla Motors Club (TMC) was the first independent online Tesla community. Today it remains the largest and most dynamic community of Tesla enthusiasts. Learn more.

Do you value your experience at TMC? Consider becoming a Supporting Member of Tesla Motors Club. As a thank you for your contribution, you'll get nearly no ads in the Community and Groups sections. Additional perks are available depending on the level of contribution. Please visit the Account Upgrades page for more details.


SUPPORT TMC
Top