Modeling human-autonomy team steering behavior in shared-autonomy driving scenarios

Well-accepted models such as the two-point steering model and its variations describe human steering behavior in non-autonomous vehicles. However, these models may not describe human steering in a shared autonomous vehicle, where the human driver cooperates with an autonomous controller. This work explores how the generalized two-point steering model, a variation of the classical two-point model, may apply to human steering in a shared autonomous vehicle. This study reports two key findings: (1) We find that humans do not necessarily steer the vehicle to the exact lane center, perhaps due to imprecise distance perception or a preference to stay off-center in the lane. Thus, we propose adding a steering bias term to the generalized steering model to account for this behavior; (2) We also find that human steering adapts so that the overall team steering–the combined human and autonomous steering input–behaves according to the generalized steering model with this new bias term. We collected data over 150 runs across 5 drivers and 3 levels of autonomy, and found that the modified generalized steering model accurately predicts team steering behavior.