To be fair/clear that is the outcome of not ignoring things on the shoulder like it used to. Which dovetails into folks allowing the cars to run into emergency vehicles.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
The Car Thread ... !
- Thread starter D.T.
- Start date
-
- Tags
- car
- Sort by reaction score
I would imagine they are minimizing the messaging to keep the drivers' eyes on the road, rather than the UI. It did say why it was changing lanes in the FSDs UI.
I still stand that this seems to be an issue that is only widespread with Tesla. That is my issue with the technology. It doesn’t feel fully-baked.
Phantom braking being a full on slam on the brakes was well before the issue of Tesla's slamming into emergency vehicles caught public attention.
Tesla should scrap the basic AP stack and just let everyone use the FSD highway stack. We don't get phantom braking anymore.
It has been a long time for me, so I presumed the braking issues cropped up more when they started trying to figure out how to not run into cross traffic stuff. the full FSD stack doesn't slam on brakes. Our complaint now is that the speed setting is a max setting not a you will go this fast setting.
This still highlights that the car doesn't give you enough information to understand why it is doing something.
As I understand it, the reason they stopped using input from Radar was specifically to address the "phantom braking." The story has it that this is what was causing the conflicts, Radar would pickup an object like a wall off to the side, or some other object and the system would panic and assume it was an object that would cause an accident. This was in combination with manual coding for AP/FSD. Now with the NN, maybe able to make the call between radar and vision without Phantom Braking.
Seen too many wham bam Tesla videos. I drive expecting the car in front to slam in their brakes.
I have wondered about the future of SuperCruise. I may be mistaken, but it seems like in its current conception, it’s a bit of a dead end. The system is dependent on streets getting LIDAR mapped by specialized vehicles. It’s just not possible for GM to map every street in America and frequently enough to keep everything updated. Plus driving off the highway is far more complicated- it has to account for street signs, stop lights, pedestrians, bicyclists, vehicles doing the wrong thing, etc.
It would presumably need a lot of visual processing and prediction advancement Tesla has made with FSD. But I suspect unlike Tesla they would develop LIDAR over optical cameras as that seems to be the safer bet, as far as I can tell.
Basically everyone else is betting on premapped roads for autonomy.
It still isn't clear to me if LIDAR can read stoplights and street signs, I thought it was purely a "more robust" version of radar.
It can't. Lidar is used for live mapping. Even companies that have Lidar are using vision for almost everything, and Lidar for checks and balances (basically to confirm what vision sees). Including Lidar is the future, it just needs to be cheap enough that it can be included in entry level vehicles like how back up cameras are now. But vision based is not going anywhere and is a problem that every manufacture is going to have to solve. Lidar is just an additional safety backup.
In my opinion it will be much easier to add Lidar after vs trying to solve vision & Lidar at the same time. The reason is, you will be so far behind the game (waiting for affordable long lasting Lidar) and will become over reliant on Lidar instead of using it only as a backup. Processing, interpretation and decision making all have a cost. Also, these Lidar sensors need to be as reliable as cameras too.
