Safety Implications of Variability in Autonomous Driving Assist Alerting
IEEE Transactions on Intelligent Transportation Systems - Aug 2022
Lots to discuss here:
TL;DR
Mary (Missy) Cummings has co-authored a research paper into whether there is any difference between three slightly different Tesla Model 3 cars in a series of Advanced Driving Assist Systems (ADAS) driving tests, mainly looking at driver monitoring.
Long story short, the cars did not perform the same as each other.
Interesting Point
L2+ systems are not regulated because ... these systems are assistive and not required for systems to operate.
NHTSA... will not intervene until it determines that some sort of defect exists that poses an unreasonable safety risk, which then leads to a recall
Her introduction and background chapters are worth the read.
- Many in the industry use the term L2+ for these type of cars (this will annoy lots of TMC people I'm sure)
- no US regulations addressing performance standards for ... L2 + systems
- lack of performance standards for known computer vision problems
- no requirements that over-the-air software updates be (formally) vetted
- NHTSA... does not address driver-assist features
- drivers ... struggle to remain engaged as autonomy increases
- There are no requirements... for proof of safe operation of L2 + systems, including driver monitoring
She chose Tesla only because "Tesla models have L2+ ADAS systems called Autopilot that can be used on interstates, divided highways and urban and rural roads...they were the only test platform available that could perform across the range of defined scenarios."
I saw some errors, some interesting points, and the driving tests hardly investigate very much. Her point is that the systems are pretty complicated and need to be studied more, especially in how driver monitoring works.
Last edited:
