Not At All. This suggests to me that the engineers hacked together enough functionality for a very controlled demo. Working in the real world with dashed freeway line markers, often painted out, obscured, in shadows, irregular due to construction, etc. is a very different problem. I know that it is very foolish to bet against Elon, but my gut tells me that his autonomous driving goal will end up being much harder than he anticipated. And for us, that means long drawn out releases with less functionality than promised.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
-
On today's TMC Podcast, we ask the question "What is the Tesla Cybercab?". Join us on YouTube live at 1PM and participate in the chat.
Schedule for Autopilot features
- Thread starter si123ca
- Start date
Not At All. This suggests to me that the engineers hacked together enough functionality for a very controlled demo. Working in the real world with dashed freeway line markers, often painted out, obscured, in shadows, irregular due to construction, etc. is a very different problem. I know that it is very foolish to bet against Elon, but my gut tells me that his autonomous driving goal will end up being much harder than he anticipated. And for us, that means long drawn out releases with less functionality than promised.
Ben W
Chess Grandmaster (Supervised)
As a software developer, I can assure you that a staged demo is never a reliable indication of being "ready for release".
In my estimation, the new hardware as described is fundamentally incapable of safe "onramp to offramp" autonomous driving. Long-range rear-facing radar is a necessity to allow safe lane changes at highway speed, specifically for detecting other vehicles approaching rapidly from behind. The 16ft Ultrasound sonar is far too short-range for this. I don't think the rear optical video camera alone would have sufficient precision / reliability / visibility, and (oddly?) there was no mention of it being used in conjunction with the Autopilot.
For that matter, even driver-prompted lane changing as demonstrated in the D test drives could be highly dangerous in real life. California DMV guidelines state that the driver must signal at least five seconds before changing lanes on a freeway. But if the Tesla Autopilot begins changing lanes immediately as soon as the driver flicks the turn signal, this violates the DMV guideline and would present a clear hazard to other drivers. (I'd be leery to pass such an Autopiloted Tesla on the left, for instance.) But if the Autopilot waits five seconds before merging, then from the driver's point of view, an indeterminate amount of time passes, and then the car will begin steering spontaneously into the next lane while the driver's eyes are (presumably) focused straight ahead. If another car is rapidly approaching from behind at that moment, neither the driver nor Autopilot will detect it in time to avoid a collision.
More to the point, the driver-initiated lane change conflates two very different activities: signaling an intent to merge, and actually merging. Pressing the turn signal has always only indicated the former, not the latter. It's completely typical to signal an intent to merge when the actual merge is unsafe at that very moment. (E.g. trying to merge into a jammed lane on the 405.) Yet if the Autopilot does not detect the unsafe condition (e.g. another car rapidly approaching from behind) and tries to merge immediately, a collision is very likely, even though the driver was "correctly" merely conveying intent to merge.
mspohr
Well-Known Member
Regd the schedule, I just noticed that the WSJ report (Tesla Aims to Leapfrog Rivals - WSJ) had some info. Quoting "people familiar with the company’s plans", it says:
- "driver assistance features such as a cruise-control system ... are expected to be released this year"
- "More advanced autopilot features .... should be delivered to cars by the first half of next year"
I think that generally matches with people's guesses in this thread.
Ben W, you have some good points regd auto lane change. My guess is that, if the rear HD camera isn't sufficient during their testing, Tesla may quietly add a rear-facing radar for this in future (and make this feature dependent on that). Another possibility is to make it a two-step process (although this might be too much manual intervention for "autopilot"): blinker for turn signal as usual, then press a button (one on the steering wheel maybe) for the car to do the lane change.
- "driver assistance features such as a cruise-control system ... are expected to be released this year"
- "More advanced autopilot features .... should be delivered to cars by the first half of next year"
I think that generally matches with people's guesses in this thread.
Ben W, you have some good points regd auto lane change. My guess is that, if the rear HD camera isn't sufficient during their testing, Tesla may quietly add a rear-facing radar for this in future (and make this feature dependent on that). Another possibility is to make it a two-step process (although this might be too much manual intervention for "autopilot"): blinker for turn signal as usual, then press a button (one on the steering wheel maybe) for the car to do the lane change.
A good intuitive way to implement that would be to have the driver pull the wheel a bit into the direction of the lane to change into, rather than a button... :wink:
mspohr
Well-Known Member
Nearly ready is not ready though.
For me personally, until the feature is available and I've actually been able to play with it on a test drive, Im not gonna buy a new car based on it coming "in the future"
This is basically why I sold my 2013 P85 and bought an S550. Paying for gas sucks though
I drove a new S550 the other day and I do have to admit it really is quite a fantastic car in many respects - including the plethora of working auto driving features. In the rest of respects it is almost everything the Model S is not, and vice versa, but that's what makes both cars remarkable vehicles.
But indeed the MB version is clearly in production. Given demos and words in the past on battery swap, valet mode, others, in Tesla's track record tells me "see production to believe it" should be in order for the Tesla current and future buyers.
Last edited:
One would think that somebody would have pointed that out to them by now. Whether pushing the turn signal lever causes it to flash three times or some other number of times isn't merely a matter of preference, but of law. It doesn't matter if I think that three flashes is too few. It's a matter of it not constituting a legal signal for a lane change. Yet Tesla never saw fit to give users the ability to adjust the number of or duration of the flashes. If they can't do something so basic for the existing software, they don't seem to be paying attention to what's legal. You have to start off signalling a lane change legally. Then you can change lanes. In some states, that's defined in terms of seconds in advance. In others, it's defined in terms of feet in advance. Either way, the car has the ability to figure out how long to signal before changing lanes. They should have gotten that part right from the beginning. After they figured that out, then they could worry about the vehicle making the lane change on its own.
Actually, you (and I) have no idea what level of completeness it is at. We know (because there are pictures floating around) that they have been testing the system on the road for many months. For all we know, it might be at the release candidate stage-- or it might have been hacked together so it only works under that exact scenario.
Given that the hardware spec has been finalized and is shipping, I would guess that it is much farther along than a "hacked together" demo.
- - - Updated - - -
I'm sure that they've thought about those issues. For all we know, there is a preference somewhere in the system which allows you to adjust the time between moving the turn signal lever and the lane change taking place.
As for the lookback, the optical camera probably makes more sense at the ranges and speeds we're talking about than radar does. They don't need a great deal of precision (they get that once it gets closer)-- they just need to know that something is closing, and a rough order of magnitude.
If they wanted the system to race in NASCAR, then radar would be a good idea (or a necessity). For ordinary driving, the system (and a human driver, for that matter), should leave more margin for error.
Given the circumstances of the demo environment, they may also have shortened the time related to the turn signal lane change-- if they waited 5 seconds, they would have run out of room on that course.
Given that the hardware spec has been finalized and is shipping, I would guess that it is much farther along than a "hacked together" demo.
- - - Updated - - -
I'm sure that they've thought about those issues. For all we know, there is a preference somewhere in the system which allows you to adjust the time between moving the turn signal lever and the lane change taking place.
As for the lookback, the optical camera probably makes more sense at the ranges and speeds we're talking about than radar does. They don't need a great deal of precision (they get that once it gets closer)-- they just need to know that something is closing, and a rough order of magnitude.
If they wanted the system to race in NASCAR, then radar would be a good idea (or a necessity). For ordinary driving, the system (and a human driver, for that matter), should leave more margin for error.
Given the circumstances of the demo environment, they may also have shortened the time related to the turn signal lane change-- if they waited 5 seconds, they would have run out of room on that course.
I still wonder how much of Elon's presentation was 'planned', and how much was Elon getting a little carried away with the moment. In particular, the idea that if you have your calendar loaded, that you would be able to summon the car to your location. I see so many issues with that, beyond the technical ones. Even though this would be limited to private land for now, it requires a precision beyond a simple address. The office I'm in now has an address, but it's huge, and there's multiple entrances. There was no mention of the car 'homing in' on your cellphone, but for this to work well, it would need something other than an address.
I love the idea, and in my head I can see a bunch of Model S's aimlessly drifting around a parking lot looking for their owners, but to me this particular feature, more than any other, seems so far off into the future.
I love the idea, and in my head I can see a bunch of Model S's aimlessly drifting around a parking lot looking for their owners, but to me this particular feature, more than any other, seems so far off into the future.
scott jones
Member
The road was also lined with tall plastic poles. Why did they need those there at all? It was either the main indicator the car used to navigate that short course, or a back up to prevent it from hitting the building. Regardless, roads are not lined with those
Hi, @donv,
What evidence do you have to support your assertion that "GM will have something similar in less than 2 years"?
Thanks,
Alan
What evidence do you have to support your assertion that "GM will have something similar in less than 2 years"?
Thanks,
Alan
Google "GM Super Cruise." This was the first link which popped up:
GM to Introduce Hands-Free Driving in Cadillac Model - Bloomberg
That says 2017 model year, which typically means introduced in fall of 2016-- so 2 years from now. Maybe not "less".
GM to Introduce Hands-Free Driving in Cadillac Model - Bloomberg
That says 2017 model year, which typically means introduced in fall of 2016-- so 2 years from now. Maybe not "less".
@donv,
Thanks for the pointer!
I'm not quite sure how to evaluate this article. It's advertising features that are two years out. Does GM have a good track record of delivering its promises in that timeframe? (I'm asking sincerely.)
I think we're all agreeing that what GM is touting is some (large?) fraction of what Tesla recently demo'd, and promised for deployment "soon". So we're comparing a Tesla "soon" with a GM "unnamed model two years from now".
The same article also touts vehicle-to-vehicle communications in GM vehicles two years out, re traffic. Isn't that what Tesla has today, deployed?
Finally, the article quotes Barra as saying the touted vehicle will "enter a segment where we don't compete today", further leading to a sense of wonder. That's a lotsa promises.
Thanks for the pointer!
I'm not quite sure how to evaluate this article. It's advertising features that are two years out. Does GM have a good track record of delivering its promises in that timeframe? (I'm asking sincerely.)
I think we're all agreeing that what GM is touting is some (large?) fraction of what Tesla recently demo'd, and promised for deployment "soon". So we're comparing a Tesla "soon" with a GM "unnamed model two years from now".
The same article also touts vehicle-to-vehicle communications in GM vehicles two years out, re traffic. Isn't that what Tesla has today, deployed?
Finally, the article quotes Barra as saying the touted vehicle will "enter a segment where we don't compete today", further leading to a sense of wonder. That's a lotsa promises.
Ben W
Chess Grandmaster (Supervised)
OMG! Call NASA now and tell them their Space Shuttle O-rings might fail catastrophically in cold weather! I'm sure they haven't thought of that!
(Shuttle blows up...)
Tesla is good but not infallible. It's deceptively easy to get tunnel vision on a large project and lose track of some important subtle detail. (Or if each person assumes someone else has taken care of it.) For the case in point, it's possible that the rear camera provides good enough telemetry that they can depend on it for this task, but that's far from obvious (particularly in poor lighting conditions), and odd that they didn't mention it at all as part of the Autopilot system.
R²B
All Star
Personally I think they're going to coincide with the "D" deliveries, I'm hoping for February 2015-ish.
There's a P85 test drive video floating around where the TM employee tells the driver that there is currently only 2/10 autopilot features implemented (lane departure warning, speed assist) and the remainder of the features are coming around February.
There's a P85 test drive video floating around where the TM employee tells the driver that there is currently only 2/10 autopilot features implemented (lane departure warning, speed assist) and the remainder of the features are coming around February.
Last edited:
it would be awesome if Tesla releases a video with self parking demo. I then know that all they need is further testing and maybe some liability release signature from customers.
http://youtu.be/b_m8DqTlOLE
http://youtu.be/b_m8DqTlOLE
Similar threads
