Perceive, Attend, and Drive: Learning Spatial Attention for Safe Self-Driving
This work addresses safety and interpretability issues in autonomous driving, though it is incremental as it extends attention mechanisms from perception to motion planning.
The paper tackles the problem of improving safety in self-driving systems by introducing a sparse attention module that learns to focus on important regions for motion planning, resulting in enhanced planner safety and interpretability.
In this paper, we propose an end-to-end self-driving network featuring a sparse attention module that learns to automatically attend to important regions of the input. The attention module specifically targets motion planning, whereas prior literature only applied attention in perception tasks. Learning an attention mask directly targeted for motion planning significantly improves the planner safety by performing more focused computation. Furthermore, visualizing the attention improves interpretability of end-to-end self-driving.