Uploaded March 2023 | Updated September 2026, 1 hour ago
Distinguished Robotics Colloquium: Human interactions with autonomous vehicles: Calibrating trust
Dawn M. Tilbury
Inaugural Chair of the Robotics Department
University of Michigan
Even as vehicles become more automated with technology advances, they will still need to interact with humans. Both humans inside the vehicle, who may need to take over the driving, and humans outside the vehicle, such as pedestrians crossing the street, must be considered. How much humans trust the automated vehicles (AVs), by making themselves vulnerable to the AVs’ actions, can affect safety and performance. If drivers overtrust the AV’s capabilities, the risks of system failures or accidents increase. On the other hand, if drivers undertrust the AV, they will not fully leverage the benefits of the AV’s functionalities. Therefore, both types of trust miscalibrations (under- and overtrust) are undesirable. We consider the problem of maintaining drivers’ trust in an automated vehicle at a calibrated level—in real time, while they are operating the vehicle. We also describe how pedestrians interact with AVs while crossing the street, and what factors affect their trust. We develop models that can predict pedestrian trajectories that can be used by the AV controller to plan safe paths. Results from human subject experiments (with simulated AVs) will be presented. A discussion of future research directions and practical considerations will conclude the talk.
For more information on the Maryland Robotics Center see:
https://robotics.umd.edu
Distinguished Robotics Colloquium: Human interactions with autonomous vehicles: Calibrating trust
Dawn M. Tilbury
Inaugural Chair of the Robotics Department
University of Michigan
Even as vehicles become more automated with technology advances, they will still need to interact with humans. Both humans inside the vehicle, who may need to take over the driving, and humans outside the vehicle, such as pedestrians crossing the street, must be considered. How much humans trust the automated vehicles (AVs), by making themselves vulnerable to the AVs’ actions, can affect safety and performance. If drivers overtrust the AV’s capabilities, the risks of system failures or accidents increase. On the other hand, if drivers undertrust the AV, they will not fully leverage the benefits of the AV’s functionalities. Therefore, both types of trust miscalibrations (under- and overtrust) are undesirable. We consider the problem of maintaining drivers’ trust in an automated vehicle at a calibrated level—in real time, while they are operating the vehicle. We also describe how pedestrians interact with AVs while crossing the street, and what factors affect their trust. We develop models that can predict pedestrian trajectories that can be used by the AV controller to plan safe paths. Results from human subject experiments (with simulated AVs) will be presented. A discussion of future research directions and practical considerations will conclude the talk.
For more information on the Maryland Robotics Center see:
https://robotics.umd.edu
