Human Side of Tesla Autopilot

The updated and revised version of the paper will be posted soon. This work is part of the MIT-AVT large-scale naturalistic driving study:

We propose a measure of “functional vigilance” that conceptualizes vigilance when drivers are allowed to self-regulate by choosing when and where to leverage the capabilities of automation and when to perform the driving task manually:

The central observations in this work is that drivers in this dataset use Autopilot for 34.8% of their driven miles, and yet appear to maintain a relatively high degree of functional vigilance. These observations are based on annotation of 18,928 disengagements of Autopilot that quantify the ability of drivers to respond to challenging driving situations during AI-assisted driving. We discuss the limitations and implications of this work in detail in the paper, however, it is important to re-state here that these findings:

  1. Cannot be directly used to infer safety as a much larger dataset would be required for crash-based statistical analysis of risk,
  2. May not be generalizable to a population of drivers nor Autopilot versions outside our dataset,
  3. Do not include challenging scenarios that did not lead to Autopilot disengagement,
  4. Are based on human-annotation of critical signals,
  5. Do not imply that driver attention management systems are not potentially highly beneficial additions to the functional vigilance framework for the purpose of encouraging the driver to remain appropriately attentive to the road.

The authors are highly cognizant that there are significant nuances in the design, analysis, and interpretation of this work. It is our hope that it will encourage serious discussion and further investigation of how seemingly subtle features of AI-assisted system design and implementation may influence the extent to which humans are able to sustain appropriate collaborative engagement with such technology.


Support for this work was provided by the Advanced Vehicle Technology (AVT) consortium at MIT. The views and conclusions being expressed are those of the authors and may not necessarily represent those of individual sponsoring organizations. All authors listed as affiliated with MIT contributed to the work only during their time at MIT as employees or visiting graduate students.

The authors would like to thank Charles Green, Jeffrey Blecher, and many anonymous reviewers for helpful feedback and suggestions through the many iterations of the work. In addition, we would like to thank the entire team at MIT behind the development and support of the MIT-AVT Study.

Contact: For questions/comments about the work, contact Lex Fridman at