Slashdot Mirror
Volkswagen Diesel Scandal Logistics Imply Sizable Conspiracy
Guinnessy writes with an interesting analysis of the Volkswagen software cheating scandal: Physics Today's Charles Day takes a look at how diesel engines work, and why it's clear it's not just a lone software engineer who came up with the cheat. "...[S]oftware is impotent without hardware. To recognize when a car was being tested and not driven, the defeat device required data from a range of sensors -- sensors that a noncheating car might not need.... Whereas it's conceivable that a single software engineer, directed by a single manager, could have secretly written and uploaded the code that ran the defeat device, installing its associated hardware would require a larger and more diverse team of conspirators," he says.
The linked article makes the point that the sensors and hardware would not be necessary. I think the writer seriously underestimates to what extent a modern car with protection systems will try to juggle different constraints. Things like non-driving wheel rotation (defeated by being on a lab stand) are needed for breaking systems and possibly to some extent to moderate throttle control for stability. Wheel movement patterns are also needed and useful, even if you don't actually have electric power steering.
Regulating the exhaust gas recirculation somehow also makes sense. You might go totally on and off, but you would certainly want to keep it at a sensible level. You want good acceleration and full combustion of fuel while still not emitting to much nitrous oxides. It makes total sense to me that you might want to design your control system to try to judge not only the current emission levels, but also the overall driving pattern (steady straight ahead, repeated stop and go, etc) with some kind of state machine to try to find the best EGR regulation regime. This requires sensors and ways to regulate the feature.
My most innocent guess about how something such as this might have happened was an intent to find a good regime that would give nice bursty performance, while keeping nitrous oxides low overall. Progressively, the control regime was pushed until it ended up in the corner where the case of EGR being properly activated under real-world conditions basically does not happen. Some parts of it might even in the end be a bug between the intended state transitions and the actual ones. Like all bugs that give performance that seem too good to be true on the metrics you really care about (fuel consumption and enjoyable driving), no-one investigated.
Do I think it happened this way? It's hard to say. Probably not. But, in one way, it's even more frightening than an evil conspiracy. It's easy to say "I wouldn't take part in a conspiracy by my employer". It's harder to say "I would never be pressed to write code with goals that could not be fulfilled, eventually find a hack that seemed to work, and maybe ignore investigating why it worked so well"...
Anti-lock brakes, computer controlled transmission shifting, variable assist power steering, fly-by-wire throttle and closed loop engine management all require sensors. Taken together, those sensors exceed what's needed to explain VW's cars ability to distinguish between active driving and a steady state test.