4 ms·

While a non-programmer might think "geez, it should have stopped, it was going to hit that truck!", I can only think how complex the system must be to detect an

While a non-programmer might think "geez, it should have stopped, it was going to hit that truck!", I can only think how complex the system must be to detect an oncoming collision and brake only when necessary. I'm sure they could tune it to be more sensitive to oncoming objects, but I'd rather defer to the drive in those circumstances..

dagw 16y ago
The story I've read was that it wasn't a problem with their the collision code as such, but that the car battery died completely shortly before the demo and the computer systems didn't start up right when the battery got recharged.
- gvb 16y ago
  Draining the battery could be argued as "human error", but it is a pretty common error. There is no "human error" in the "computer systems didn't start up right" half of the excuse. That is a design error. Having said that, I've worked with avionics. Avionics power systems are absolutely horrible with respect to voltage regulation and glitches. The power source is switched between a ground cart, APU, and multiple engine generators at various times and causes glitches each time. Creating hardware and software that is able to start reliably when hit by multiple glitches with random intervals and random duration plus power brownouts and surges is extremely difficult. We used a programmable power supply that was able to do random glitches (it was affectionately known as the "widow maker" because the engineers' wives would become a virtual widows while the engineers worked out the bugs ;-). When the systems engineers would assure us we would have clean power, guaranteed, we would just roll our eyes and roll over the widow maker... (The battery bus actually is reliable and glitchless, but only flight critical hardware gets to hook up to that.) Just to complicate life, power glitch testing (and brake system safety testing) is an instance of the halting problem. http://en.wikipedia.org/wiki/Halting_problem http://en.wikipedia.org/wiki/Halting_problem
  - Retric 16y ago
    IMO, there are more fundamental problems. Run any computer system long enough and you will get random bit errors in both memory and computation. Avionics systems mitigate this though redundant systems but cars are not built to that level of safety. So yes, assisted braking systems will fail. The real question is how often and to what result.
    - timthorn 16y ago
      I am led to understand that ABS systems often do have multiple independent microcontrollers.
      - JshWright 16y ago
        I know of very few cars with multiple rotation speed sensors on the wheels (and the system has to know the rotation speed of each wheel if the system is to function properly). So each sensor (and all the associated cabling, wire harnesses, etc) is a single point of failure. ABS systems are far from fail-proof...
  - ComputerGuru 16y ago
    Not sure how the glitch testing is an instance of the halting problem, could you expand on that a bit?
    - gvb 16y ago
      Testing by experimentation (random glitch spacing and random glitch duration) is inherently an instance of the halting problem. If you don't have any failures, you know all the combinations you have tested work, but you can never be sure there isn't a combination still out there that you didn't hit that would cause a failure.

dagw 16y ago

The story I've read was that it wasn't a problem with their the collision code as such, but that the car battery died completely shortly before the demo and the computer systems didn't start up right when the battery got recharged.

gvb 16y ago
Draining the battery could be argued as "human error", but it is a pretty common error. There is no "human error" in the "computer systems didn't start up right" half of the excuse. That is a design error. Having said that, I've worked with avionics. Avionics power systems are absolutely horrible with respect to voltage regulation and glitches. The power source is switched between a ground cart, APU, and multiple engine generators at various times and causes glitches each time. Creating hardware and software that is able to start reliably when hit by multiple glitches with random intervals and random duration plus power brownouts and surges is extremely difficult. We used a programmable power supply that was able to do random glitches (it was affectionately known as the "widow maker" because the engineers' wives would become a virtual widows while the engineers worked out the bugs ;-). When the systems engineers would assure us we would have clean power, guaranteed, we would just roll our eyes and roll over the widow maker... (The battery bus actually is reliable and glitchless, but only flight critical hardware gets to hook up to that.) Just to complicate life, power glitch testing (and brake system safety testing) is an instance of the halting problem. http://en.wikipedia.org/wiki/Halting_problem http://en.wikipedia.org/wiki/Halting_problem
- Retric 16y ago
  IMO, there are more fundamental problems. Run any computer system long enough and you will get random bit errors in both memory and computation. Avionics systems mitigate this though redundant systems but cars are not built to that level of safety. So yes, assisted braking systems will fail. The real question is how often and to what result.
  - timthorn 16y ago
    I am led to understand that ABS systems often do have multiple independent microcontrollers.
    - JshWright 16y ago
      I know of very few cars with multiple rotation speed sensors on the wheels (and the system has to know the rotation speed of each wheel if the system is to function properly). So each sensor (and all the associated cabling, wire harnesses, etc) is a single point of failure. ABS systems are far from fail-proof...
- ComputerGuru 16y ago
  Not sure how the glitch testing is an instance of the halting problem, could you expand on that a bit?
  - gvb 16y ago
    Testing by experimentation (random glitch spacing and random glitch duration) is inherently an instance of the halting problem. If you don't have any failures, you know all the combinations you have tested work, but you can never be sure there isn't a combination still out there that you didn't hit that would cause a failure.