Description
End-to-end driving models must handle overtaking, intersections, reversing and more, and one model rarely masters all. DriveMoE from SJTU brings mixture-of-experts into vision-language-action driving models, routing scenarios to specialized experts.
It also releases the Drive-π0 baseline, published at CVPR 2026.
MoE:Scenario specialists.
Vision experts:Camera selection.
Action experts:Driving behaviors.
It also releases the Drive-π0 baseline, published at CVPR 2026.
Features
MoE:Scenario specialists.
Vision experts:Camera selection.
Action experts:Driving behaviors.
Screenshots
Tags:autonomous-driving