Presentation
Using Inverse Reinforcement Learning to Determine the Factors Contributing to the Visual Demand of Driving
SessionWednesday Poster Session
DescriptionWhat is too distracting or how long it takes to take over from an automation failure or many other risk-related aspects of driving depend upon the demands of driving, its difficulty. To drive, one needs to see, so a simple method to assess the demands of driving is to ask a driver to close their eyes whenever they can, with the percentage of times their eyes are open being the demand. To provide proper experimental control, we do this in a driving simulator, and have the driver press the button to show the road scene, visible for half a second after each button press.
To develop such predictions, we had16 drivers drive 4 test courses (2 ovals, 2 random sequences of straight and curve sections) multiple times. The data were somewhat “messy” (for example, participants did not drive at a constant speed as requested) so we used a machine learning method known asInverse Reinforcement Learning to analyze the data. This method allowed us to separate out the effects of the speed driven, curves, and other factors in the predictions of visual demand, something that could have not been done easily using traditional analysis methods.
To develop such predictions, we had16 drivers drive 4 test courses (2 ovals, 2 random sequences of straight and curve sections) multiple times. The data were somewhat “messy” (for example, participants did not drive at a constant speed as requested) so we used a machine learning method known asInverse Reinforcement Learning to analyze the data. This method allowed us to separate out the effects of the speed driven, curves, and other factors in the predictions of visual demand, something that could have not been done easily using traditional analysis methods.
Event Type
Poster
TimeWednesday, October 21st5:30pm - 6:30pm PDT
Location
Similar Presentations