161–170 of 495 posts
Earlier quoted context omitted.
Do you understand the difference between going in a straight line and going at a consistent speed? Cruise control only enables the latter.
To go straight in a car, just release the steering wheel.
Earlier quoted context omitted.
To play devil's advocate: Assuming self-driving cars are connected to each other, either directly or via the internet, if one car crashes, or even just makes a minor mistake, the knowledge of how to avoid that can be transmitted to every other car on the planet.
Seems unlikely that the knowledge of how to avoid making a mistake could be compiled without human analysis. The technology to do that autonomously doesn’t exist. It would almost require artificial general intelligence, and if we had that self driving would be trivial.
Earlier quoted context omitted.
This sounds good, but is impractical. First, a human who is paying attention makes better decision than a human who is surprised by being given control during a crisis. Therefore if automation sees a crisis, it's standard has to be that it is worse than a surprised human, and not how it compares to an alert driver. Second, the analysis that concludes that we're now in a specific situation where a human would do bette…
This is called the Halting Problem. We don't waste time trying to make machines that do that because it's theoretically undecidable. Furthermore, you're then throwing more neural network (that has to be trained improvably across the dataset which we have no proof is the full dataset required to fully enumerate the problem space), to successfully detect the problem areas in the data going into the driving NN for edge…
It is also entirely wrong.
The problem that I am describing is not the Halting Problem. A theoretically perfect NN is also able to do a lot better than any person could. The ability to have binocular vision in multiple directions, and mix that with radar awareness of the local environment and electronic communication with other cars to be aware of obstacles that are out of sight. This technology can drive more quickly and safely than humans ever could.
This sort of thing is admittedly some time off. But it is doable, and it is almost certain to happpen within our lifetimes.
Earlier quoted context omitted.
By that argument, every car with cruise control (i.e. almost all of them) has autopilot. Clearly, with their naming of their system as autopilot rather than cruise control, Tesla is trying to imply that autopilot means something more. Namely, it means the driver / pilot can offload the work of routine driving / flying to the computer, only intervening in certain times (e.g. off0highway or takeoff / landing) or in the…
Do you understand the difference between going in a straight line and going at a consistent speed? Cruise control only enables the latter.
Earlier quoted context omitted.
If the system is even a few percent better than humans then it would save thousands of lives when widely deployed. Musk has made this argument many times. Basically don't let perfect be the enemy of better. But if the vendor is legally liable in each case then this software is untenable.
Tesla (and/or Musk) keep saying it's safe to let the car drive, and then when you let the car drive and it kills you, they say you're holding it wrong. If you're actually trying to increase safety, the right things to build are computers to supervise the human driver, and not systems that require a human to supervise the computer driver. If the human drifts out of the lane, nudge the car back in and beep, but don't p…
Earlier quoted context omitted.
Stereo cameras produce line of sight depth information That is wrong. Stereo cameras provide data which, after significant processing, can result in line-of-sight depth information. With radar (including LIDAR) and ultrasound, the depth information is essentially "free" since very little processing is required to extract the depth information (i.e., the information returned is the depth). So the biggest difference be…
Aren't the algorithms for extracting depth from stereo imagery already solved and thus also essentially "free"?
The problem with Tesla's "Autopilot" is that it's marketed heavily as just that: autopilot. It's good enough to work that way 99% of the time, too, which is a recipe for disaster: things that work 99% of the time but have the potential to fail catastrophically often get glossed over by humans because it's hard to keep paying attention to something that rarely fails.
The lane markers in this area, based on video I watched of it years ago, are pretty bad, IIRC. Like, a human could have problems following the markers bad. As evidence to support this I'll say that there had been an accident in this exact area days (again, IIRC) before that destroyed the "crumple zone" in this area. :-)
It's going to be extremely interesting to see how the courts rule in a case where "I knew the Tesla on autopilot would drive in to the barrier in this area, and I turned on autopilot in this area and it drove into the barrier."
Earlier quoted context omitted.
If the system is even a few percent better than humans then it would save thousands of lives when widely deployed. Musk has made this argument many times. Basically don't let perfect be the enemy of better. But if the vendor is legally liable in each case then this software is untenable.
Humans can be 30+% better than "humans" by 1. not using a phone and 2. not being drunk or stoned. A few percent better than "humans" is quite bad relative to a semi-responsible human driver.
Yes it should be (imho) at least an order of magnitude better than humans before we consider it OK to not pay attention.
Not a few percentage points better as you say.
However, to overcome media drama, it probably will need to be not just one, but more like three or four orders of magnitude safer than humans, because of the “oh look, a Tesla had an accident” effect.
Earlier quoted context omitted.
Stereo cameras produce line of sight depth information That is wrong. Stereo cameras provide data which, after significant processing, can result in line-of-sight depth information. With radar (including LIDAR) and ultrasound, the depth information is essentially "free" since very little processing is required to extract the depth information (i.e., the information returned is the depth). So the biggest difference be…
Aren't the algorithms for extracting depth from stereo imagery already solved and thus also essentially "free"?
Radar, LIDAR, ultrasound: returning packets have X value, where X is run through a single algorithm to return depth. In the case of ultrasound and radar, the strength of the return packet is generally directly proportional to depth (algorithmically). LIDAR systems might require an additional processing step depending on the type of LIDAR system used, since many systems use shifted-arrays or rotating frequencies.
Stereovision: returns two matrices of data (i.e., each image). Each matrix is run through edge-detection and object-detection algorithms, and the resultant data is then run through more algorithms to match and compare objects between the two images. Finally, once objects have been matched up, a parallax algorithm determines their 3d positioning. While this can all be done in hardware (i.e., ASICS), it's still multiple additional levels of processing that must be done before you can even get the data to the decision-making part of the system. Stereovision is thus always inherently laggier than a radar-based system when it comes to determining depth.
Earlier quoted context omitted.
That's a very common perception, but it's not true! Kalman filters have for years been used to get usable data out of these and multiple studies have shown that ultrasound can be used as a standalone in low-speed scenarios. Design and experimental study of an ultrasonic sensor system for lateral collision avoidance at low speeds - https://ieeexplore.ieee.org/abstract/document/1336460 Environmental perception and mult…
Kalman filters over what state space? Localization isn't really the challenging aspect of self-driving cars.