Slashdot Mirror
Washington Finds Computer Simulation Unreliable
Toadpipe writes "Washington State Court of Appeals reverses a conviction in which a computer simulation had been the main evidence. Quoting 'At issue was PC-Crash, a computer program distributed by Vancouver, B.C.-based MacInnis Engineering Associates. The program recreates traffic collisions using simulations and reconstructions.
"PC-Crash had not been validated for the purpose for which the evidence was offered, simulation and prediction of multiple-occupant movement within a vehicle during a multiple-collision accident," the Court of Appeals said in ordering a new trial. "There is no general acceptance in the relevant scientific community of the use of the PC-Crash program for the purposes to which it was put."' Here is
the Court's opinion."
The Senate already informed us of that long ago.
I read the court transcript because it was quite interesting.
... well, you get the point.
Turned out the simulation that was generated that kinda matched what happened had the data entered at random until it matched "Heusser manipulated data by entering arbitrary 'inputs' such as separation speeds as high as 1,114.8 mph, placing the mailbox pole away from where it was actually located, and having the computer occupant models remain in a default resting position after the collision with the mailbox".
Indeed, the software was described at the end of the trial as "During closing argument to the jury, the prosecutor described PC-CRASH as a
computer program that essentially takes the laws of physics and reduces them to mathematical calculations that can be done over and over again to generate an accurate picture of what happened during a collision based on the tire marks at the scene, based on the physical evidence in the case such as the damage to the car, as well as the conditions that can be observed at the scene.
13 Report of Proceedings at 13. The prosecutor then showed the PC-CRASH
video to the jury, again.
Sipin was convicted as charged."
Whilst the expert opinion from someone who used his brain to see what happened described something completely different. The jury was mislead to how good the software was quite clearly, they were lead to believe that the software was infallible. It is only as good as the person entering the data, and when they choose to ignore data because it is inconvenient
So whilst the guy was stupid for buying a manual car when he had gout and couldn't drive it half the time, he does deserve a retrial.
This may be shocking, but I am actually familiar with both this software and the process of giving expert testimony. PC-Crash is one of several *Crash* programs provided by different vendors that share a common lineage. It and its sister programs are used extensively in accident reconstruction and the results are presented to juries every day. The core of these *Crash* programs are a series of well-established (although certainly not perfect) algorithms and physical properties related to vehicle dynamics. The problem here was the extension to occupant dynamics, not the use of simulation programs in general.
You may now return to your regular uniformed ranting.
You cannot, even in theory, predict how a human with arms and legs banging around will move in a complex crash. It's chaotic, in the formal mathematical sense of the world. That is, an arbitrarily small change in the initial conditions can create a large change in the outcome. In Falling Bodies, if you change the low order digit of a double precision number in the initial conditions for a fall down a staircase, the simulations will start to diverge after about a second, and the fall may end quite differently.
I had this discussion a few years ago with an Army officer who was trying to reduce accidents in parachute landings, and was considering using Falling Bodies. I talked him out of it.
Auto collisions can be simulated well because there's one big mass that dominates the simulation. So you get a deterministic result within some error limits. Multibody systems with joints and links are quite different.
Realistically, you can probably do a sound simulation which predicts how a passenger will bounce around from the beginning of the collision to the first passenger interior collision with the vehicle. Beyond that point, forget it.