Pedestrian detection

[Cnixon]

Last night my car warned me of a pedestrian,dressed in black, in an unlit area. Impressed as I had not seen the bloke wandering in the road. Now the not so good bit. Loud chimes look down at screen (is that a good thing?) Not much to see then i saw the vague outline of a person on the screen, dark red on black screen, hardly visible. The colouring of objects red during the day is fine but at night maybe another colour or at least a lighter red. I cannot recall if there was any braking. It is quite a regular occurence at the moment. 2020.12.11.1 software!

 

[Roy W.]

Not so good, as in loud chimes, look down at screen, and THUD!

 

[Alex987854]

and THUD!

Second audible warning

 

[Mr H]

Not so good, as in loud chimes, look down at screen, and THUD!

I've found that works great with the MAMIL's

 

[phillcom3]

I wish mine would show me bikes more It never shows them as I pass them same with cars on the side. If it showed parked cars I'd feel much more confident in autopilot

 

[bhav]

I wish mine would show me bikes more It never shows them as I pass them same with cars on the side. If it showed parked cars I'd feel much more confident in autopilot

Why are you trying to use it on streets with parked cars?

"Traffic-Aware Cruise Control is primarily
intended for driving on dry, straight roads,
such as highways and freeways. It should not
be used on city streets."

 

[phillcom3]

Why are you trying to use it on streets with parked cars?

"Traffic-Aware Cruise Control is primarily
intended for driving on dry, straight roads,
such as highways and freeways. It should not
be used on city streets."

Because these are straight roads and quite wide. It should eb able to by now due to all litriture deal with a normal wide town road which has a few cars down the side which it does however the question wasn't why can't i drive the road but

Why does the parked cars not show or any car to either side?

 

[bhav]

Because these are straight roads and quite wide. It should eb able to by now due to all litriture deal with a normal wide town road which has a few cars down the side which it does however the question wasn't why can't i drive the road but

Why does the parked cars not show or any car to either side?

Because it's not designed for either of those conditions? I'm not sure I understand why it "should" be able to handle things like that by now...it's not designed to and given that it's not a surprise that showing parked cars is not a priority for the visual.

In fact, if they bring in the visual it will only encourage people to use autopilot even more on roads it's not designed for.

By the way this is no surprise - as soon as you have parked cars/stationary objects it adds so many variables that could cause things to go wrong that we're just not there yet with the technology. I know Tesla's is more advanced than others on the cameras/visuals but it's still nowhere near to what Elon claims it to be (safely).

Something worth reading: Why emergency braking systems sometimes hit parked cars and lane dividers

 

[MrBadger]

Because these are straight roads and quite wide. It should eb able to by now due to all litriture deal with a normal wide town road which has a few cars down the side which it does however the question wasn't why can't i drive the road but

Simple

You don't have full self drive and even if you did, its a future feature

 

[spon88]

FSD or not, many of us will be very surprised if automatic driving on city streets is ever realised with the current set of hardware. At least to a point where you could feel confident and safe

 

[MrBadger]

Which part of the hardware do you think is deficient?

Until the autopilot rewrite is complete, I personally think it is too early to tell. If it not up to it at that point, Tesla will have no where to hide.

I do feel though that V2I capability will be needed medium term, but that will be software and an upgrade to the LTE radio I suspect. Lidar is I believe a short term fix only and Tesla will crack the vision/radar based approach. It is clear that Tesla currently have still some way to go, but they have not yet exhausted the techniques available them with current hardware set. The autopilot rewrite will be a big step, but still further improvements will be available, they will not have exhausted the capabilities of current hardware. If they ever embrace V2I (there are things possible with that which a car cannot solve on its own no matter what sensors it has), that will also be a big step likewise some form of HD mapping, the latter I am not a fan of because it has serious practical limitations. That will be the point when FSD should be capable of meeting its objectives, but whether it will or not will be up to Tesla to accept that embracing the way that others do things would also bring benefits.

I think at some point, Tesla will be forced to adjust how they detect driver awareness and possibly geofence self drive capabilities, the former may require internal camera capabilities to be modified - there are far better ways than steering wheel torque sensing. B pillar cameras will also likely get a demister (although autopilot rewrite and redundancy it brings may make misting of these cameras less of an issue), likewise a heater for front radar for extreme conditions. The only sensor missing that I would like is rear facing radar (along with upgrade to forward facing radar capabilities). So apart from missing rear facing radar (which isn't strictly necessary), the missing capabilities would be relatively inexpensive upgrades.

imho.

 

[spon88]

Which part of the hardware do you think is deficient?

Until the autopilot rewrite is complete, I personally think it is too early to tell. If it not up to it at that point, Tesla will have no where to hide.

I do feel though that V2I capability will be needed medium term, but that will be software and an upgrade to the LTE radio I suspect. Lidar is I believe a short term fix only and Tesla will crack the vision/radar based approach. It is clear that Tesla currently have still some way to go, but they have not yet exhausted the techniques available them with current hardware set. The autopilot rewrite will be a big step, but still further improvements will be available, they will not have exhausted the capabilities of current hardware. If they ever embrace V2I (there are things possible with that which a car cannot solve on its own no matter what sensors it has), that will also be a big step likewise some form of HD mapping, the latter I am not a fan of because it has serious practical limitations. That will be the point when FSD should be capable of meeting its objectives, but whether it will or not will be up to Tesla to accept that embracing the way that others do things would also bring benefits.

I think at some point, Tesla will be forced to adjust how they detect driver awareness and possibly geofence self drive capabilities, the former may require internal camera capabilities to be modified - there are far better ways than steering wheel torque sensing. B pillar cameras will also likely get a demister (although autopilot rewrite and redundancy it brings may make misting of these cameras less of an issue), likewise a heater for front radar for extreme conditions. The only sensor missing that I would like is rear facing radar (along with upgrade to forward facing radar capabilities). So apart from missing rear facing radar (which isn't strictly necessary), the missing capabilities would be relatively inexpensive upgrades.

imho.

I'm no expert but in an earlier life I did a lot of work in process control. What I learned is that if the primary instrumentation couldn't cut it, then no matter what level of fancy interpretive software you hung off the back end, you still wouldn't get the result you desired. This was exacerbated as the process became more sophisticated.

So basic concerns such as camera fogging and rain concern me. Yes, you could retrofit equipment to improve that and maybe the full software re-write will also help a lot. But a long way to go I think it's fair to say. For what it's worth I didn't buy in to FSD specifically for those reasons but I have been in cars where its been in use and it was unimpressive, actually a bit scary. I will use AP infrequently on the motorway but still wary of odd ball braking and occasional weirdness at off ramps etc. All that said, I prefer to just drive it!

 

[bhav]

I think at some point, Tesla will be forced to adjust how they detect driver awareness and possibly geofence self drive capabilities, the former may require internal camera capabilities to be modified - there are far better ways than steering wheel torque sensing. B pillar cameras will also likely get a demister (although autopilot rewrite and redundancy it brings may make misting of these cameras less of an issue), likewise a heater for front radar for extreme conditions. The only sensor missing that I would like is rear facing radar (along with upgrade to forward facing radar capabilities). So apart from missing rear facing radar (which isn't strictly necessary), the missing capabilities would be relatively inexpensive upgrades.

imho.

Yes, in my view the only realistic actual "FSD" in the short/medium term (for me that's next 10 years) is one which is geofenced and has driver awareness detection.

 

[Godsense]

Yes, in my view the only realistic actual "FSD" in the short/medium term (for me that's next 10 years) is one which is geofenced and has driver awareness detection.

Manufacturers do need to work on the driver detection system. I completely disagree with the way people try and bypass the AP nag. To the point there should be points and a fine for using a defeat device. Though I guess the police could get you for driving without due care?

 

[MrBadger]

The rest of the industry and academia seem pretty certain that internal camera monitoring the driver is the answer. Not sure on the capabilities of the Model 3 camera which is not on the S or X, but I can see Tesla being a bit reluctant going that route. Where I use to work, we use to run a GoPro to monitor driver awareness on our simulator so can be done with a single driver facing camera.

 

[bhav]

I believe the technical issue (in general, not Tesla's view) is that it doesn't work very well in low light/darkness - so the false positives/false negatives could undermine its use and cause other unintended consequences. I'm sure the tech will improve though, or there will be some other method.

 

[MrBadger]

Our simulator was very dark - the only light came from the 300 degree projection screen and internal instrumentation. The driver face images I saw seem to have no issues picking out the drivers eyes. Plus also, its not all about what you can see.

 

[Battpower]

Until the autopilot rewrite is complete, I personally think it is too early to tell.

Annoyingly, the 'rewrite' makes any discussion about current features and performance moot for HW3 cars (as they are hoping for better things to come).

However, what is still valid is new niggles / glitches being introduced to current cars. I am finding excessive and unnecessary 'take control now' warnings - with no automated controls engaged - quite distracting and often when you least need to be distracted. It already cost me a sped camera flash (genuinely) by beeping and flashing on a dark unfamiliar road just leaving a roindabout as I needed to have been looking into the dark for a speed limit sign. (40 in 30 mph limit)

Yesterday, on country road with no center lines, passing cyclists and got loads of beeps and warnings. I've had similar before, but still distracting.

Also, pedestrians seem to randomly trigger warnings. Not sure, but it could be linked to if the road has centre marks.

 

[MrBadger]

What version are you on? What you describe inflicted various versions of 2020.8, but I thought that issue had been largely resolved now.

 

[Battpower]

I wish mine would show me bikes more It never shows them as I pass them same with cars on the side. If it showed parked cars I'd feel much more confident in autopilot

I imagine that showing parked cars is always going to be tough. Deciding if stationary cars, especially if they are parked over hanging the carriageway, must be very difficult to handle. Is it a queue waiting to turn left? Is a car pulling out of a parking space? Or in the process of parking?