Perhaps you got here after reading my previous post entitled Road Trip! In that post I describe a situation where taking over from FSD in emergency situations can be a problem. And if you’re here because you want to know what the heck OOTL is, then welcome.
So, here’s the problem. you’re ‘driving’ down the road with Tesla’s Full Self Driving (FSD) enabled. You’re doing 55 because that’s what the the construction zone requires. You have two men moving temporary caution cones across the roadway, one just clearing the roadway and the other just starting to move across the roadway. The car begins to slow, but there’s traffic behind you, and you’re just not sure if FSD is going to navigate the situation properly so you take over. That’s the moment the Out-of-Loop (OOTL) preformance problem begins.
The content included below was provided by Gemini using this prompt:
Let’s talk about the psychology of ‘supervised full self driving’ cars. Is it true that the driver is at a disadvantage in emergency situations due to the fact that the driver is monitoring edge cases while the car is in FSD mode and not comprehending the full situation prior to an emergency? If this is the case, then the driver’s response would have incomplete information to properly execute an emergency maneuver. Is there credible research in this area that indicates that this could be a problem? Also an issue is the time required to disable FSD prior to responding to the emergency situation manually.
Response:
There has been decades of research in cognitive psychology and ergonomics that document the phenomenon that human factors researchers call the “Out-of-the-Loop” (OOTL) performance problem, and it is one of the most critical safety vulnerabilities in “supervised” or Level 2 semi-autonomous driving systems (such as Tesla’s Autopilot or Full Self-Driving) (Gouraud et al., 2017; Nordhoff et al., 2023).
Credible peer-reviewed research strongly indicates that relying on a human driver to instantly resolve an emergency while a vehicle is driving itself places that human at a severe cognitive disadvantage.
1. The Vigilance Decrement and “Edge Case” Monitoring
You noted that the driver is forced to monitor edge cases rather than comprehending the full situation. Psychologically, human beings are fundamentally ill-suited for this task.
When a car drives smoothly for long stretches, the brain naturally shifts into a state of complacency and overtrust (Nordhoff et al., 2023). Because the manual workload drops, the human mind enters what is called mind wandering (MW), decouples its focus from the immediate external environment, and decreases cognitive comprehension of traffic dynamics (Gouraud et al., 2017).
Instead of actively reading the road—predicting the behavior of a pedestrian, checking blind spots, or feeling the traction of the asphalt—the driver becomes a passive monitor. When an emergency edge case suddenly occurs, the driver is forced to jump from low mental engagement to maximum emergency response instantly.
2. Incomplete Information & Situation Awareness (SA)
Your point about the driver lacking full context to properly execute a maneuver is highly accurate. Cognitive psychologist Mica Endsley’s foundational framework breaks Situation Awareness (SA) into three levels:
- Perception: Seeing the elements around you.
- Comprehension: Understanding what those elements mean right now.
- Projection: Anticipating what will happen in the next few seconds.
When Supervised FSD is active, the car is handling the Comprehension and Projection loops. When the system makes a sudden error or encounters an unhandled edgecase, the human experiences a “surprise effect” or “mode confusion” (Gouraud et al., 2017; Nordhoff et al., 2023).
Because the driver was “out of the loop,” they do not possess Level 2 or Level 3 SA. They might look up, perceive an obstacle (Level 1), but lack the critical fraction of a second needed to understand why the car is moving toward it or where the safest escape path lies. This leads to erratic, panicked physical inputs, such as over-correcting the steering wheel or slamming the brakes inappropriately.
3. The Time Tax: Disengagement and Latency
The time required to comprehend a situation, disable the system, and take manual action is a massive bottleneck. In human factors research, this sequence is studied as the takeover time.
In standard manual driving, human perception-reaction time to a hazard (like brake lights ahead) is roughly 1.0 to 1.5 seconds. However, when transitioning from automated control back to manual control during a failure, that timeline expands significantly:
- The Takeover Delay: Studies tracking unstructured or sudden transitions of control show that it typically takes a driver anywhere from 2.0 to over 5.0 seconds to safely re-orient, take physical control, and navigate a road hazard safely (Mok et al., 2015).
- The “2-Second” Danger Zone: Research measuring critical transitions indicates that when a driver is forced to take over with only 2 seconds of warning, the majority fail to safely avoid the hazard (Mok et al., 2015). They require closer to 5 to 8 seconds of lead time to perform at the same level as a traditional manual driver (Mok et al., 2015).
- The Physical and Mental Disengage Tax: To disable FSD, a driver must either apply physical torque to the steering wheel, press the brake, or tap a stalk. While the physical act takes a fraction of a second, the cognitive hurdle—deciding whether to fight the car’s steering or trust it for one more second—causes a measurable cognitive delay.
The “Irony of Automation”
First coined by Lisanne Bainbridge in 1983, this classic psychological paradox states that the more advanced an automated system becomes, the more crucial yet unreliable the human supervisor becomes (Nordhoff et al., 2023). By leaving the human only with the task of fixing the system when it fails, we ask them to perform a highly complex task (emergency response) precisely when they are least mentally prepared to do so.
FSD systems actually double the driver’s burden in complex environments. Instead of just tracking the road, the driver has to track the road AND what the AI thinks about the road (Nordhoff et al., 2023). This causes cognitive fatigue, making sudden, split-second manual interventions inherently prone to incomplete data execution.
References
- Gouraud, J., Delorme, A., & Berberian, B. (2017). Autopilot, Mind Wandering, and the Out of the Loop Performance Problem. Frontiers in Neuroscience, 11, Article 541. https://doi.org/10.3389/fnins.2017.00541
- Mok, B., Johns, M., Lee, K. J., Miller, D., Sirkin, D., Ive, P., & Ju, W. (2015). Emergency, Automation Off: Unstructured Transition Timing for Distracted Drivers of Automated Vehicles. 2015 IEEE 18th International Conference on Intelligent Transportation Systems, 2458–2464. https://doi.org/10.1109/itsc.2015.396
- Nordhoff, S., Lee, J. D., Calvert, S. C., Berge, S., Hagenzieker, M., & Happee, R. (2023). (Mis-)use of standard Autopilot and Full Self-Driving (FSD) Beta: Results from interviews with users of Tesla’s FSD Beta. Frontiers in Psychology, 14, Article 1101520. https://doi.org/10.3389/fpsyg.2023.1101520