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

Dangerous bug with autopilot on 2018.10.4

This site may earn commission on affiliate links.
LIDAR doesn't do anything especially impressive either, I tend to believe Musk is on the right pat in spite of what everyone else is saying. That having been said, my experience with LIDAR is fairly limited, so it could be that I've only dealt with poor implementations.

I wonder if the big 3 (and maybe others) pushed for the SAE to include LIDAR in their FSD standard as a barrier of entry. It wouldn't be that hard to believe, really. Just like it's not that hard to believe that the big three sponsor at least some lobbying to force electric vehicles to make sound for "safety."

I’m not a student of vehicle autonomy but Waymo does seem to have the reputation for having the most advanced system for the time being.

The problem is the pricing and having us all drive cars that look like Ecto-1.

Chevy super cruise has lidar on one control car that transmit data to rest of fleet.

Easily accomplishable by Tesla as well with their well connected cars backed by space x low orbit low latency satellites.

Tesla waze seems like child play with the possibilities.
 
I find it also interesting that the headlights turn on in the first video, and are already on in the second. Is this just an auto feature of the wipers coming on? Also, the Road surface around the centerline looks very un-standard.
May be related to some state laws that require headlights on when using wipers. My 2015 produced Volt turns on lights after 15 seconds of wipers. I’m in MA and they added this law about a year ago.
 
They can do all of the software updates possible... is this not disconcerting when Tesla says the vehicles have all the hardware necessary for fully autonomous self driving?
I'm a broken record on this, but I feel the Lidar issue is not going to go away.
Would the car be better or worse if it had Lidar?
If the answer is better, then it seems like a necessity.
I disagree (but don’t want to hit disagree lest you think I’m angry):) if hills and crests can be solved, so can this.
 
Also regarding the thread post, I don’t think this is a dangerous bug, it’s not ideal and certainly will be fixed but dangerous may not be the right word l, if it happened on FSD then it would be dangerous ... certainly potentially dangerous , but this thread seems like a blader thread....
 
They can do all of the software updates possible... is this not disconcerting when Tesla says the vehicles have all the hardware necessary for fully autonomous self driving?
I'm a broken record on this, but I feel the Lidar issue is not going to go away.
Would the car be better or worse if it had Lidar?
If the answer is better, then it seems like a necessity.

Why would Lidar help in this case? The only advantage of Lidar is its ability to provide distance information - this might help to detect object on or near the road but I can't see how that information helps you to find the center of a lane.

The information of precise distance of every point around you is mostly noise. For self-driving I don’t think it provides anything in principle that can’t be achieved with stereo cameras. The key is really AI vision – if you have it then it doesn’t matter if the data is provided by camera or Lidar, and if you don’t have it then you have to hard-code algorithms to handle various conditions – in these cases Lidar data might be easier to use because it gives you distance directly, but long term hard-coded algorithms is a dead end so at the end Lidar doesn't really buy you much.
 
Why would Lidar help in this case? The only advantage of Lidar is its ability to provide distance information - this might help to detect object on or near the road but I can't see how that information helps you to find the center of a lane.

The information of precise distance of every point around you is mostly noise. For self-driving I don’t think it provides anything in principle that can’t be achieved with stereo cameras. The key is really AI vision – if you have it then it doesn’t matter if the data is provided by camera or Lidar, and if you don’t have it then you have to hard-code algorithms to handle various conditions – in these cases Lidar data might be easier to use because it gives you distance directly, but long term hard-coded algorithms is a dead end so at the end Lidar doesn't really buy you much.

Fantastic first post.

Problem is how much patience does the public have with AI vision

and

do we have the hardware to execute AI FSD as is on the 2.0/2.5 platform..
 
Problem is how much patience does the public have with AI vision

and

do we have the hardware to execute AI FSD as is on the 2.0/2.5 platform..
I am not an expert on self-driving, but my guess is that there are two problems to solve here: 1. recognizing various objects/elements (lane, sign, people, car, etc.) and their positions/speeds/accelerations, etc. 2. Deciding how to drive once given a set of such elements/objects. The first problem probably can be solved pretty efficiently by neuro nets and my guess is that current 2.0/2.5 hardware is sufficient. The second problem I think can be more efficiently solved by rule based systems (with a relatively small set of rules) than neuro nets so the current hardware should be able to handle as well. Rule based system probably will not be able to cover all corner cases so it might not be sufficient for L5 but I feel that it should be able to cover enough cases for L2-4.

The videos posted by various people on how the new release handles some pretty tricky traffic conditions shows that tsla is pretty close to solving these problems so my guess is they will be able to get to L4 in the not too distant future.
 
IMO, when cars are L5 ready, we still need steering wheel so that we can drive the car manually when some of the sensors are broken/dirty and AP stops working and remote control also not possible. We see all the time people driving car even when the muffler is broken, one headlight not functional etc..

I have only seen cars that have both manual as well as sensor based control for wipers. Akin to what any auto manufacturer did with wipers, auto steering should follow the same.
 
Last edited:
Hello,

I just made an account specifically because I wanted to post here and say that I experienced this exact same issue twice over the weekend, one time coming painfully close to swerving directly into a car in the next lane over.

Unfortunately I can't confirm the firmware version as it was just a rental Model X that I no longer have, but my assumption would be that it is kept up to date.

Everything was fine when I used a little wiper fluid (a normal amount), but the windscreen was absolutely caked with bugs after a drive through a national park in Wales, so I really pumped that sucker to try and clean them off. It seems that using an excessive amount of wiper fluid caused the autopilot to freak out, as the first time I did this the car drifted into the middle lane of the motorway, much like your video.

I'd forgotten this incident when I deployed a whole heap of wiper fluid for a second time to try and clean them off again (12 hour drives will do that to you...), and we jerked so forcefully into the middle lane that we were physically thrown around in our seats. A car was screaming past us in that middle lane at the exact same moment - maybe the car was adjusting after the initial swerve to avoid collision, which could be why it was so incredibly abrupt the second time.

I know I'm not really adding any additional evidence or bankable intel on the issue, but I noticed that no one else in the thread really seemed to be able to replicate the issue so I just wanted to say that you're not alone!