Portfolio: VehiCal

Modeling Attention in Semi-Autonomous Vehicles

"Self-driving" is kind of a misnomer.

Autonomous and semi-autonomous systems can encounter situations where it does not know what to do. Human intervention is required to ensure physical safety. timely attention of a person is required to take over some aspect of decision making or control.

But what can we do to get the attention of a likely distracted driver to ensure a safe outcome?

I built the system and ran the ~50 person user study that served as the basis for this experiment using Amazon Mechanical Turk

Using the results of the experiment, we were able to formalize the perception hand-off problem using a Partially Observable Markov Decision Process (POMDP) model with a problem specific structure. Further details can be found in the paper.

We demonstrate that the state of attentiveness does indeed impact the human performance in a statistically significant way. Our proposed intervention method resulted in 5.5% more corrective responses