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

Do I dare : Firmware 6.2

This site may earn commission on affiliate links.
Curious considering the relatively recent comment saying the hardware wouldn't be significantly upgraded in that area for the next 2 years. (Or something like that.)

Not that curious because Telsa is not trying to do autonomous urban driving. They're trying to do onramp to offramp driving. TACC can do some urban driving but it's not super great at it. Here's the problems I've seen with it. Sometimes it'll see objects positioned very near to the lane as cars in the lane and will harshly brake trying to avoid them. I see this behavior most commonly with temporary road signs or those standing signs that businesses position by the road. Cars moving in and out of your lane cause it to slow down and then not resume for quite a bit of time. Example if you have a car that turns off into a driveway. It slows down but fairly quickly gets out of your way. TACC will slow down pretty dramatically, often coming almost to a complete stop and then won't resume for quite a bit of time. Out on the highway TACC needs very little micromanaging. But in town I find myself overriding it quite a bit. I've noticed this because I drive a highway that goes through towns and drops as low as 25 mph while most of the time it's 55 mph so I leave TACC running and just adjust it up and down.
 
Not that curious because Telsa is not trying to do autonomous urban driving. They're trying to do onramp to offramp driving. TACC can do some urban driving but it's not super great at it.

Anyone who thinks the current Tesla hardware is close to being capable of autonomous city driving has another thing coming... Google's autonomous cars have $150,000 in equipment including a 64 beam Velodyne LIDAR range finder mounted on top. It builds a real-time 3d map of its environment and processes it using neural network AI. Evenso, if you check out the wikipedia article on it, (https://en.wikipedia.org/wiki/Google_driverless_car) the latest Google prototype cannot "handle heavy rain and snow-covered roads", there are also other limitations on discerning objects such as trash and debris that can unnecessarily veer the vehicle, the lidar technology cannot spot potholes or humans signaling the car to stop, are unable to recognize temporary traffic signals, have not proven themselves in snow or rain, are unable to navigate through parking lots, are unable to differentiate between pedestrian and policeman or between crumpled up paper and a rock.

If Google's $150k LIDAR system is so limited, I think it's safe to say that our Model Ses simply do not have anywhere close to the brute force computing power, sensor stack or software smarts to do full-on autonomous city driving.

Now, putting you in a safe, optimal weather highway cruise with lane holding at 75mph is totally within reason for the current hardware and the software to do that should be quite refined within the next year or so. Furthermore, i imagine that pulling out of your garage to meet you at the curb type situations are also well within reach with the current hardware, otherwise they wouldn't have promised it, but I think that's about where it's gonna end as far as the current hardware stack goes.
 
Last edited:
Now, putting you in a safe, optimal weather highway cruise with lane holding at 75mph is totally within reason for the current hardware and the software to do that should be quite refined within the next year or so. Furthermore, i imagine that pulling out of your garage to meet you at the curb type situations are also well within reach with the current hardware, otherwise they wouldn't have promised it, but I think that's about where it's gonna end as far as the current hardware stack goes.

This is all I want, because it takes care of 90% of the driving that I do. I'd feel safer leaving the urban driving to myself. I also don't imagine the garage thing being that difficult to implement. I'd just record the human input for entering and exiting the garage. And then have the car repeat the human input while keeping the camera and sensors on to make sure no kid randomly jumps in front of the car.
 
I had a chance to speak to a coworker doing research into autonomous driving (I love my job) and she pointed out to me that current state of the art assumption is that for truly autonomous driving in a crowded environment (i.e., inner city) you need at least half a dozen cameras for this to work. So I'm not that surprised that Elon expects to need more sensors...
The current state of the art for humans is two relatively poor resolution optical sensors mounted very close together with limited gimbaling, both on the same swivel mount, and it works well enough. Some humans have just one working optical sensor and still drive safely. So we have a working example where the number of sensors can be as few as one optical input and still perform the task...that implies to me it isn't more sensors that are needed, but rather more processing of what the sensors are reporting.
 
The current state of the art for humans is two relatively poor resolution optical sensors mounted very close together with limited gimbaling, both on the same swivel mount, and it works well enough. Some humans have just one working optical sensor and still drive safely. So we have a working example where the number of sensors can be as few as one optical input and still perform the task...that implies to me it isn't more sensors that are needed, but rather more processing of what the sensors are reporting.

Of course when a car does it by itself, it will need to be 100 times safer to be accepted. And that one sensor is very mobile - I don't envision autonomous cars as having mobile sensors. So an eyeball equivalent in 4 directions at least. And of course the Model S does not have this.