Human Brain vs TACC

[Wingsy]

The other day my wife and I went to the store. We were on a rural road, wide and well marked. An oncoming car had slowed to a stop or nearly stopped. Then it turned left, crossed our path and entered a driveway. It was far enough away that had it stopped in the middle of my lane I would have had time to stop without using maximum braking force. I let it go without braking at all and I did this pretty much unconsciously. This is what I think my computer (brain) did in this situation:

* Oncoming vehicle detected in left lane ahead, stopped or nearly stopped. Unusual, so dedicate an additional 200,000 neurons to monitor this.
* Vehicle advanced into my lane and is headed straight for me. Path of vehicle is curved and projected path and speed will take it out of a collision threat if it continued. Watch closely.
* Computed time to collision if vehicle stops in my lane is 4.8 seconds. Maximum braking time for full stop to avoid collision is 3.2 seconds. Waited 1 second before applying brakes to see what vehicle does.
* Vehicle continued along its projected curved path, still partially in my lane but now the vehicle cannot stop in my lane due to its forward speed and its reasonable braking power.
* Continued driving without braking.

There was no collision or threat of collision and passenger remained calm. If I was using TACC it would have panicked at the first sign that the oncoming vehicle was in my lane (dumping that bag of groceries in the back seat and prompting a scream into my right ear), regardless if the vehicle was on a path that would take it out of a collision threat within time for my car to brake to a full stop. I'm not saying "if I can do this why can't Tesla", but I am puzzled as to why this seems to be so difficult. Tesla does so many other things that require some heavy computin' so why not this? It's got the required information to do so.

Just wondering.

 

[jjrandorin]

The other day my wife and I went to the store. We were on a rural road, wide and well marked. An oncoming car had slowed to a stop or nearly stopped. Then it turned left, crossed our path and entered a driveway. It was far enough away that had it stopped in the middle of my lane I would have had time to stop without using maximum braking force. I let it go without braking at all and I did this pretty much unconsciously. This is what I think my computer (brain) did in this situation:

* Oncoming vehicle detected in left lane ahead, stopped or nearly stopped. Unusual, so dedicate an additional 200,000 neurons to monitor this.
* Vehicle advanced into my lane and is headed straight for me. Path of vehicle is curved and projected path and speed will take it out of a collision threat if it continued. Watch closely.
* Computed time to collision if vehicle stops in my lane is 4.8 seconds. Maximum braking time for full stop to avoid collision is 3.2 seconds. Waited 1 second before applying brakes to see what vehicle does.
* Vehicle continued along its projected curved path, still partially in my lane but now the vehicle cannot stop in my lane due to its forward speed and its reasonable braking power.
* Continued driving without braking.

There was no collision or threat of collision and passenger remained calm. If I was using TACC it would have panicked at the first sign that the oncoming vehicle was in my lane (dumping that bag of groceries in the back seat and prompting a scream into my right ear), regardless if the vehicle was on a path that would take it out of a collision threat within time for my car to brake to a full stop. I'm not saying "if I can do this why can't Tesla", but I am puzzled as to why this seems to be so difficult. Tesla does so many other things that require some heavy computin' so why not this? It's got the required information to do so.

Just wondering.[/QUOTE]

Just wondering what.... Why we (as humans) can evaluate the above scenario better than a computer? Computers have a hard time telling the difference between things like an inflatable car and a real one, while we humans would not have such trouble. The reason computers are "better" drivers than humans is because we dont pay attention (at least not 100%). If all humans devoted 100% of their concentration to driving when they are driving, Its my belief that we would be better than any computer by a LOONG shot.

We dont, however, even when by ourselves, and even when NOT fiddling with phones, and music and such. We "think about stuff" when we drive... we listen to audio books, we talk to other humans... and thats all before we get to the people who eat things that require utensils while driving.

True story.. just two days ago, while my wife and I were taking a trip from Temecula to San Marcos to look at smokers in a BBQ store, on the freeway, whie, traffic was going 80 miles an hour, we saw a man, in a minivan, with 2 kids in car seats in the back, "driving" with a fast food cup in his left hand, and a burger in his right hand. He basically had his wrists on the steering wheel, in an older minivan, driving 2 kids, at 80+ miles an hour.

THATS why computers are better drivers than us right now, because there is always someone on the road doing something like that. Personally, I believe you would WANT the car to err on the side of caution. Would you rather 100 times spilled groceries and noise, or 1 time when the car "made a mistake" and didnt take action when it should have?

 

[bwilson4web]

The other day my wife and I went to the store. We were on a rural road, wide and well marked. An oncoming car had slowed to a stop or nearly stopped. Then it turned left, crossed our path and entered a driveway. It was far enough away that had it stopped in the middle of my lane I . . .

Consider it a lesson on how to drive on basic autopilot.

Bob Wiilson

 

[jyalpert]

The other day my wife and I went to the store. We were on a rural road, wide and well marked. An oncoming car had slowed to a stop or nearly stopped. Then it turned left, crossed our path and entered a driveway. It was far enough away that had it stopped in the middle of my lane I would have had time to stop without using maximum braking force. I let it go without braking at all and I did this pretty much unconsciously. This is what I think my computer (brain) did in this situation:

* Oncoming vehicle detected in left lane ahead, stopped or nearly stopped. Unusual, so dedicate an additional 200,000 neurons to monitor this.
* Vehicle advanced into my lane and is headed straight for me. Path of vehicle is curved and projected path and speed will take it out of a collision threat if it continued. Watch closely.
* Computed time to collision if vehicle stops in my lane is 4.8 seconds. Maximum braking time for full stop to avoid collision is 3.2 seconds. Waited 1 second before applying brakes to see what vehicle does.
* Vehicle continued along its projected curved path, still partially in my lane but now the vehicle cannot stop in my lane due to its forward speed and its reasonable braking power.
* Continued driving without braking.

There was no collision or threat of collision and passenger remained calm. If I was using TACC it would have panicked at the first sign that the oncoming vehicle was in my lane (dumping that bag of groceries in the back seat and prompting a scream into my right ear), regardless if the vehicle was on a path that would take it out of a collision threat within time for my car to brake to a full stop. I'm not saying "if I can do this why can't Tesla", but I am puzzled as to why this seems to be so difficult. Tesla does so many other things that require some heavy computin' so why not this? It's got the required information to do so.

Just wondering.

This is an incredibly complex set of problems to solve. You've got to:

- create a 3d map of your surroundings
- identify all objects of interest in that map
- forecast the future state of the 3d space many time steps ahead, including your own relative position to all identified objects. And by the way, this probably involves an estimation of the intent of each object of interest and a probability of various actions the object might take, so multiple forecasts may be required for likely outcomes.
- decide if anything in your forecast is a risk
- if a risk is identified, you have to take an action
- the potential space for taking action is large so you have to optimize the action to the particular circumstance

And then you have to do this for a really hugely wide array of conditions. The ability to handle this well is really, really hard and there are tons of different strategies for breaking down and solving the problem.

Contrast this with a much simpler task: follow a car in front of you at a distance of 100ft. Then what you have to do is:

- identify if there is a car in front of you
- compute the distance to the car
- subtract the distance from 100ft, call that the error
- make a speed adjustment that is proportional to the error

 

[jjrandorin]

Hmm.. my reply got embedded in OPs quote, so I am going to break it out here as its not clear from my original post exactly which words are OPs and which are mine... so below is my original post in this thread:

=====================================

Just wondering what.... Why we (as humans) can evaluate the above scenario better than a computer? Computers have a hard time telling the difference between things like an inflatable car and a real one, while we humans would not have such trouble. The reason computers are "better" drivers than humans is because we dont pay attention (at least not 100%). If all humans devoted 100% of their concentration to driving when they are driving, Its my belief that we would be better than any computer by a LOONG shot.

We dont, however, even when by ourselves, and even when NOT fiddling with phones, and music and such. We "think about stuff" when we drive... we listen to audio books, we talk to other humans... and thats all before we get to the people who eat things that require utensils while driving.

True story.. just two days ago, while my wife and I were taking a trip from Temecula to San Marcos to look at smokers in a BBQ store, on the freeway, whie, traffic was going 80 miles an hour, we saw a man, in a minivan, with 2 kids in car seats in the back, "driving" with a fast food cup in his left hand, and a burger in his right hand. He basically had his wrists on the steering wheel, in an older minivan, driving 2 kids, at 80+ miles an hour.

THATS why computers are better drivers than us right now, because there is always someone on the road doing something like that. Personally, I believe you would WANT the car to err on the side of caution. Would you rather 100 times spilled groceries and noise, or 1 time when the car "made a mistake" and didnt take action when it should have?

 

[BEPA400]

The other day my wife and I went to the store. We were on a rural road, wide and well marked. An oncoming car had slowed to a stop or nearly stopped. Then it turned left, crossed our path and entered a driveway. It was far enough away that had it stopped in the middle of my lane I would have had time to stop without using maximum braking force. I let it go without braking at all and I did this pretty much unconsciously. This is what I think my computer (brain) did in this situation:

* Oncoming vehicle detected in left lane ahead, stopped or nearly stopped. Unusual, so dedicate an additional 200,000 neurons to monitor this.
* Vehicle advanced into my lane and is headed straight for me. Path of vehicle is curved and projected path and speed will take it out of a collision threat if it continued. Watch closely.
* Computed time to collision if vehicle stops in my lane is 4.8 seconds. Maximum braking time for full stop to avoid collision is 3.2 seconds. Waited 1 second before applying brakes to see what vehicle does.
* Vehicle continued along its projected curved path, still partially in my lane but now the vehicle cannot stop in my lane due to its forward speed and its reasonable braking power.
* Continued driving without braking.

There was no collision or threat of collision and passenger remained calm. If I was using TACC it would have panicked at the first sign that the oncoming vehicle was in my lane (dumping that bag of groceries in the back seat and prompting a scream into my right ear), regardless if the vehicle was on a path that would take it out of a collision threat within time for my car to brake to a full stop. I'm not saying "if I can do this why can't Tesla", but I am puzzled as to why this seems to be so difficult. Tesla does so many other things that require some heavy computin' so why not this? It's got the required information to do so.

Just wondering.

That’s why we have a frunk. There is even the little hook thing to keep your groceries upright. That way you can let TACC react like a maniac. I don’t have an answer for the scream in your right ear but if you figure that out please let me know. I get that even without hard braking.