Discuss: Should Tesla Adopt NVIDIA’s Cooperative Steering? [VIDEO]

[Not a Tesla App]

NVIDIA's Alpamayo platform introduces Cooperative Steering, a feature that blends human input with autonomous control. Should Tesla adopt this approach for FSD?… Read More

Should Tesla Adopt NVIDIA’s Cooperative Steering? [VIDEO]

NVIDIA's Alpamayo platform introduces Cooperative Steering, a feature that blends human input with autonomous control. Should Tesla adopt this approach for FSD?

www.notateslaapp.com

 

[TESLASUPERFAN]

I’ve said this for over 2 years, COLLABORATION MODE IS SALIENT with hazardous road conditions. Grok needs access to alert u while FSD IS ENGAGED to TEMPORARILY TAKE OVER. FSD then RECORDS your actions(how u handle the scenario). Next, FSD reengages, analyzes the exact steps u took then stores that info to be used when similar scenario appears in the future. Grok should say “please take over for a moment” screen border glows red indicating MANUAL DRIVE MODE…..after driver takes over, averts obstacle, grok says “FSD reengaged, thank u”-screen border glows green or blue.

 

[SoMEtesla]

I’ve said this for over 2 years, COLLABORATION MODE IS SALIENT with hazardous road conditions. Grok needs access to alert u while FSD IS ENGAGED to TEMPORARILY TAKE OVER. FSD then RECORDS your actions(how u handle the scenario). Next, FSD reengages, analyzes the exact steps u took then stores that info to be used when similar scenario appears in the future. Grok should say “please take over for a moment” screen border glows red indicating MANUAL DRIVE MODE…..after driver takes over, averts obstacle, grok says “FSD reengaged, thank u”-screen border glows green or blue.

Grok isn't running on the car, FSD is and it would be what asks for help. I think what NVIDIA does is different than you describe. They let you control the car by applying slight forces on the steering wheel and don't seem to make a fuss when you do. There are at least two well documented cases where Tesla's FSD was accidently overridden by the driver resulting in an accident. In one case the driver was unaware they had brushed up against the steering wheel but the data shows the wheel was turned. In the other the driver was reaching into the back seat area and their leg pushed hard enough against the wheel to disengage FSD. In an aircraft the pilots follow a verbal positive transfer of control protocol "You have the controls", "I have the controls", "You have the controls." To do otherwise invites FSD has needed to allow us to take over because it is still learning to drive. At some point letting us take over by moving the controls will make the drive less safe. In some cases it already is. I've had the car aggressively decelerate and shift lanes at highway speed to avoid a car that started pulling into my car. Not knowing the cause of the abrupt movement I would have overridden the steering wheel movement with a negative result. NVIDIA's system would seemingly require much less force than FSD to override it. It may make sense at this point in its education but it seems problematic in the long run. Dammed if you do, dammed if you don't....

 

[TESLASUPERFAN]

Grok isn't running on the car, FSD is and it would be what asks for help. I think what NVIDIA does is different than you describe. They let you control the car by applying slight forces on the steering wheel and don't seem to make a fuss when you do. There are at least two well documented cases where Tesla's FSD was accidently overridden by the driver resulting in an accident. In one case the driver was unaware they had brushed up against the steering wheel but the data shows the wheel was turned. In the other the driver was reaching into the back seat area and their leg pushed hard enough against the wheel to disengage FSD. In an aircraft the pilots follow a verbal positive transfer of control protocol "You have the controls", "I have the controls", "You have the controls." To do otherwise invites FSD has needed to allow us to take over because it is still learning to drive. At some point letting us take over by moving the controls will make the drive less safe. In some cases it already is. I've had the car aggressively decelerate and shift lanes at highway speed to avoid a car that started pulling into my car. Not knowing the cause of the abrupt movement I would have overridden the steering wheel movement with a negative result. NVIDIA's system would seemingly require much less force than FSD to override it. It may make sense at this point in its education but it seems problematic in the long run. Dammed if you do, dammed if you don't....

TOTALLY DISAGREE-1000 scenarions where FSD ran lights, phantom braking, slowed down for no reason, flies over speed bumps, traversed into on coming traffic, etc ,etc. A simple alert (voice and visual) to TEMPORARILY take over SAVES lives, simply folding your arms when approaching HEAD ON TRAFFIC hoping FSD will correct the situation is idiotic and foolish that could cost you and others their lives. Especially when a momentary steering wheel correction would have saved lives.

We definetly DONT NEED YET ANOTHER VOICE (FSD)-nav voice and grok is already overkill. GROK absolutely needs to be intergrated with ALL the features of the car.....this is what every car manufacturer is implementing. Way overdue for AI chat bots to expand the capabilities.

Collaboration mode (at this time) is a must have feature and I assume NHTSA will enhance the laws accordingly in the future. TOTALLY relying on AI/UFSD for every scenario, especially where death is salient, extreme weather, etc is foolish and opens the gates for class action lawsuits. Pointing a blaim-thrower at FSD and shrugging your shoulders ain't gonna fly in court.

Flying a plane in clutter free open skies with an experience pilot and comparing it to driving on a 10 lane super Freeway, where there are hundreds of INEXPERIENCE drivers all around you, all going different directions, with plenty of sub-standard vehicles is so NOT AN ACCURATE COMPARISON.....not by any means. Two entirely different AI/computers.

 

[SoMEtesla]

TOTALLY DISAGREE-1000 scenarions where FSD ran lights, phantom braking, slowed down for no reason, flies over speed bumps, traversed into on coming traffic, etc ,etc. A simple alert (voice and visual) to TEMPORARILY take over SAVES lives, simply folding your arms when approaching HEAD ON TRAFFIC hoping FSD will correct the situation is idiotic and foolish that could cost you and others their lives. Especially when a momentary steering wheel correction would have saved lives.

We definetly DONT NEED YET ANOTHER VOICE (FSD)-nav voice and grok is already overkill. GROK absolutely needs to be intergrated with ALL the features of the car.....this is what every car manufacturer is implementing. Way overdue for AI chat bots to expand the capabilities.

Collaboration mode (at this time) is a must have feature and I assume NHTSA will enhance the laws accordingly in the future. TOTALLY relying on AI/UFSD for every scenario, especially where death is salient, extreme weather, etc is foolish and opens the gates for class action lawsuits. Pointing a blaim-thrower at FSD and shrugging your shoulders ain't gonna fly in court.

Flying a plane in clutter free open skies with an experience pilot and comparing it to driving on a 10 lane super Freeway, where there are hundreds of INEXPERIENCE drivers all around you, all going different directions, with plenty of sub-standard vehicles is so NOT AN ACCURATE COMPARISON.....not by any means. Two entirely different AI/computers.

You point out the problems that have been occurring while FSD expands its understanding of how to drive the car safely. Re read my reply and you should note that I acknowledge a learning period that requires control override. I also did not dismiss FSD needing the ability to request assistance which it already does. I pointed out the problems of letting the human with less information and skill override FSD's decision making when it reaches the point that it is truly FSD.

In your world the driver is fully attentive and processing everything going on around the vehicle. When FSD is really good at what it does the driver will not be aware of the full problem. In the two accidents I mentioned the drivers were not aware enough of what was going on to realize they had imparted a force sufficient enough to take control from FSD. INVIDIA plans to require very little force based on their videos. Once NVIDIA's self driving is good enough to instill driver confidence they will experience many more accidents like those tesla has. Picking the right level of ease in allowing transfer of control is a challenging problem.

My mention of pilot transfer of control was to highlight how important it is to understand who is driving. The accidents encountered in aviation due to mishandling this have nothing to do with the complexity of the "airway", they have to do with the aircraft crashing when no one is flying it. In a helicopter it can take as little as 1/2 second for the rotor system to diverge enough to strike the helicopter itself.

Back to GROK - GROK runs in a server somewhere connected to the car by an unreliable cellular data link. The latency and packet loss rate of that connection precludes its use in critical event handling. It is great for navigating and would be helpful with other non safety critical controls. The AI running in the car, controlling the driving, deciding when it needs help is FSD.

 

[TESLASUPERFAN]

Grok 5 is one of the smartest AI on the planet-touching the screen constantly and re-directing your focus while driving is dangerous. AI CHAT Bot's primary function is to interact with systems and verbal requests. Intergrating with UFSD in paramount in Attaining UFSD.

I dont hear or see FSD asking users politely, to take over mili-seconds before potential hazzard pops up- a blue flashing border with some out dated BEEPS, surely the techs can implement something user friendly.

Predictive coding, along with sharper cameras, or strategically placed additional cameras and some ever-changing clever algorithms could solve this quite easily. Resulting in potential problems (that FSD struggles with) can be avoided altogether.

When U align Grok 5 to co-exist with the UFSD-it will appear flawless, smooth, intelligent.....almost human. Buying or Riding in a vehicle without AI interaction will appear antiquated. The fusion of UFSD and the AI assistant will be the deciding factor in avoiding potential concerns BEFORE they occur. Yup, its that smart and getting smarter.

 

[Procal]

You point out the problems that have been occurring while FSD expands its understanding of how to drive the car safely. Re read my reply and you should note that I acknowledge a learning period that requires control override. I also did not dismiss FSD needing the ability to request assistance which it already does. I pointed out the problems of letting the human with less information and skill override FSD's decision making when it reaches the point that it is truly FSD.

In your world the driver is fully attentive and processing everything going on around the vehicle. When FSD is really good at what it does the driver will not be aware of the full problem. In the two accidents I mentioned the drivers were not aware enough of what was going on to realize they had imparted a force sufficient enough to take control from FSD. INVIDIA plans to require very little force based on their videos. Once NVIDIA's self driving is good enough to instill driver confidence they will experience many more accidents like those tesla has. Picking the right level of ease in allowing transfer of control is a challenging problem.

My mention of pilot transfer of control was to highlight how important it is to understand who is driving. The accidents encountered in aviation due to mishandling this have nothing to do with the complexity of the "airway", they have to do with the aircraft crashing when no one is flying it. In a helicopter it can take as little as 1/2 second for the rotor system to diverge enough to strike the helicopter itself.

Back to GROK - GROK runs in a server somewhere connected to the car by an unreliable cellular data link. The latency and packet loss rate of that connection precludes its use in critical event handling. It is great for navigating and would be helpful with other non safety critical controls. The AI running in the car, controlling the driving, deciding when it needs help is FSD.

If you're "discussing" with the person I think you are, my suggestion is, don't bother. Keep your sanity, lol.

 

[Procal]

Grok isn't running on the car, FSD is and it would be what asks for help. I think what NVIDIA does is different than you describe. They let you control the car by applying slight forces on the steering wheel and don't seem to make a fuss when you do. There are at least two well documented cases where Tesla's FSD was accidently overridden by the driver resulting in an accident. In one case the driver was unaware they had brushed up against the steering wheel but the data shows the wheel was turned. In the other the driver was reaching into the back seat area and their leg pushed hard enough against the wheel to disengage FSD. In an aircraft the pilots follow a verbal positive transfer of control protocol "You have the controls", "I have the controls", "You have the controls." To do otherwise invites FSD has needed to allow us to take over because it is still learning to drive. At some point letting us take over by moving the controls will make the drive less safe. In some cases it already is. I've had the car aggressively decelerate and shift lanes at highway speed to avoid a car that started pulling into my car. Not knowing the cause of the abrupt movement I would have overridden the steering wheel movement with a negative result. NVIDIA's system would seemingly require much less force than FSD to override it. It may make sense at this point in its education but it seems problematic in the long run. Dammed if you do, dammed if you don't....

By the way, I agree with your reasoning. Collaborative steering would have been really helpful/useful in the early FSD (beta) days, but now, I think it is more of a liability than an aid. If anything, I think Tesla just needs a more robust way to report issues without actually disengaging FSD, especially for those scenarios where it messes up, but it wasn't necessarily the "wrong" thing to do. Along the same lines, it would be awesome if they would finally implement some sort of crowdsourcing function for things like potholes, construction, speed cameras, etc. similar to Waze, Apple Maps, Google Maps.

 

[TESLASUPERFAN]

so a car heading into oncoming traffic with UFSD in control is NOT a liability, resulting in MULTIPLE deaths but a COLLABORATION where a HUMAN corrects the steering (saving multiple deaths) is a liability ?

Dont think ANY lawyer fresh out of law school, will agree with that nor do I think that will fly by NHTSA investigators. If that were even remotely true, there would be a class action lawsuit sign up sheet a mile long.

 

[SoMEtesla]

so a car heading into oncoming traffic with UFSD in control is NOT a liability, resulting in MULTIPLE deaths but a COLLABORATION where a HUMAN corrects the steering (saving multiple deaths) is a liability ?

Dont think ANY lawyer fresh out of law school, will agree with that nor do I think that will fly by NHTSA investigators. If that were even remotely true, there would be a class action lawsuit sign up sheet a mile long.

take a breath- let the shift key have a rest. If you actually had the current version of FSD you might better understand the problem. I still need to be able to override it at this point but - if I have my hands on the steering wheel ready to intervene I will probably prevent the car from saving me from potential collision more times than I will save it. It reacts quicker than you can, well before you can look around and determine why. If you overpower the wheel it is on you to deal with the problem that you may not yet be aware of. 2050 miles since 14.2 - 2005 miles of that self driving.

 

[SoMEtesla]

By the way, I agree with your reasoning. Collaborative steering would have been really helpful/useful in the early FSD (beta) days, but now, I think it is more of a liability than an aid. If anything, I think Tesla just needs a more robust way to report issues without actually disengaging FSD, especially for those scenarios where it messes up, but it wasn't necessarily the "wrong" thing to do. Along the same lines, it would be awesome if they would finally implement some sort of crowdsourcing function for things like potholes, construction, speed cameras, etc. similar to Waze, Apple Maps, Google Maps.

Thanks, agreed on the extra reporting. I think having the ability for FSD to automatically resume after detecting the driver is once again not on the controls like NVIDIA would be helpful in those situations where FSD has accidentally been overridden. Maybe even use the cabin camera to see if it was intentional (is driver reaching behind the seat and not watching the road?) The two accidents I mentioned probably would not have been avoided due to how quickly they happened. I have caused a disconnect once accidently which would have benefitted from such a behavior. Coincidentally, a video was posted yesterday on youtube of a cyber cab with only the customer in it. The customer grabbed the wheel to see what would happen. Tesla has cranked up the breakaway torque required to take control or eliminated the capability because he didn't get control.

 

[Procal]

Thanks, agreed on the extra reporting. I think having the ability for FSD to automatically resume after detecting the driver is once again not on the controls like NVIDIA would be helpful in those situations where FSD has accidentally been overridden. Maybe even use the cabin camera to see if it was intentional (is driver reaching behind the seat and not watching the road?) The two accidents I mentioned probably would not have been avoided due to how quickly they happened. I have caused a disconnect once accidently which would have benefitted from such a behavior. Coincidentally, a video was posted yesterday on youtube of a cyber cab with only the customer in it. The customer grabbed the wheel to see what would happen. Tesla has cranked up the breakaway torque required to take control or eliminated the capability because he didn't get control.

Yeah, I saw the video, and actually linked it in a separate post. I think that may actually be a good feature to implement in customer cars, but instead of making it mandatory, making it an option similar to how you can control chill vs standard or sport acceleration. At least at first, and allow people to get accustomed to it slowly. I personally haven't had any issues with accidental disengagements, but I can certainly see cases where the small amount of torque required to disengage could be an issue. Especially in the scenario you described of reaching for something in the back seat.

 

[TESLASUPERFAN]

ITS NOT just steering: it is often braking for say red lights, blowing thru stop signs that may be faded, snow or dirt covered, sun light in camera lens, dirty camera, or "ghost in the machine" I.E. techno glitches.

There are plenty of reasons UFSD might fail, stop for a moment and require human intervention. Collaboration mode should NOT be an option is should be READILY available at a moments notice. Having zero recourse when FSD screws up opens the flood gates to danger/lawsuits. Famous last words "FSD was in total control, I could not do ANYTHING to avoid a head on collision".

Again, safety comes first and everything fails once in awhile for a multitude of reasons or for no reason at all (ghost in the machine-refers to random strings of code behaving unpredictable). I think...... for a LOT of people (at this time) they would prefer to have an over-ride function in worse case scenario's.

I think NHTSA and Lawyers would all agree that a "fail-safe" is far better option that no option at all. Totally relying on AI/UFSD will be man kinds downfall...........insert any robot movie/self driving car episode you like (eyes glow red, flashing warning lights just prior to things going awry).

For the record: I don't have a HW4 nor do I have he latest version of FSD, I have not tested or researched chips like NVIDIA. I see the potential, theories, the need but I don't see the empirical evidence that this cannot fail, that it's safer than a human driver without the need for some sort of manual intervention.

I have yet to see the actual implementation, testing, research or any investigations. A lot of techno talk with big promises with zero in field testing. Lives are at stake and thats a huge gamble to endorse without some sort of backup in place.

 

[SoMEtesla]

Yay - They have implemented the take control after an accidental disconnect.
Also first steps of being more assertive with other controls when it is important.

Tesla Introduces 'Automatic Collision Evasion' for Manual Drives - Not a Tesla App

Tesla has quietly added a major new active safety capability to its vehicles, turning its autonomous driving software into an emergency crash-prevention tool

www.notateslaapp.com

 

[TESLASUPERFAN]

And I’m sure I was not credited -although reading this is totally ridiculous to lock this behind a pay subscription wall. A life saving feature for when the cars in manual mode but only if u subscribe is non sense

 

[TESLASUPERFAN]

Time to switch back to uss from Tesla vision. I rarely get any warnings with near misses from cross traffic at low speeds.

The ultra sonic sensors at least warn me.

 

[SoMEtesla]

Time to switch back to uss from Tesla vision. I rarely get any warnings with near misses from cross traffic at low speeds.

The ultra sonic sensors at least warn me.

I'm not sure I understand your point. Are you saying that you are experiencing FSD near misses where you need to take control at low speeds? Or are you driving (FSD off) and it isn't warning you when you are getting close to objects maneuvering at low speeds? 12460 of 12618 miles FSD since 14.2 and while it does maneuver closer to objects than I would at slow speed it is not hitting them. I have come to understand that means it is not too close for FSD. I agree with your other comment that auto braking should also steer regardless of FSD licensing. The complicating part is that if you haven't been using FSD regularly you won't have built the trust level needed to not override the sudden evasive maneuver. They are helping with that problem by being more forceful with the steering and other control limitations but my money says you would be startled and try to override it anyway.

 

[TESLASUPERFAN]

No Fsd-driving on parking lot, low speeds-car cut in front of me on more than one occasion and no warning whatsoever: no brakes, no red visuals, no audibles…..nothing-I slammed on my brakes.