Skip to yearly menu bar Skip to main content

Affinity Workshop: WiML Workshop 1

Reading the Road: Leveraging Meta-Learning to Learn Other Driver Behavior

Anat Kleiman · Ryan Adams


A significant challenge for autonomous vehicles today is learning to drive with other drivers. This stems from the need to accurately model others drivers' actions, which is especially difficult because of how many unique driving styles can exist. We present a meta learning approach: By treating the experience of driving with others as tasks, the deep learning model creates a unique representation for each experience. During test time, the model can quickly adapt to unseen driving styles with only a few updates. We present promising initial results: the meta learning model outperforms reinforcement learning baselines, even accounting for exposure to new drivers.

Chat is not available.