Tuesday, November 5, 2019
Tesla on-the-fly software updates
One of Tesla's major selling points is that they can, and do, make software changes on-the-fly.
In a legacy vehicle, change control is rigorously enforced so when there is an accident, forensic engineers can reconstruct the chain of failures that lead to the crash. They look at skid marks, witness marks on seat belt retractors, shiny spots on fabric, broken welds, the nature of the fractured plastic parts. They interrogate the flight recorder and collect the last thirty seconds of the vehicle's life and five seconds of post-crash data.
If something in the vehicle's design contributed to the bad-outcome, then that knowledge goes into the institutional knowledge and that mistake is not made again.
Was a bolt in the suspension to hard and therefore subject to hydrogen embrittlement? Did the carpet creep forward and trap the accelerator pedal? Was the spring in the key cylinder too short or too compliant to keep the vehicle's motor on? Those errors are captured and fixes are integrated into new designs.
And often, victims sue the automaker to recover damages.
The $64 Billion questions are: When does Tesla's quick-like-bunny, on-the-fly software changes become evidence tampering? What mechanisms exist to ensure that lessons learned in one generation are carried over into the next? As software becomes more massive, what procedures are in place to make sure the software plays nicely together?