Perceive, Attend, and Drive: Learning Spatial Attention for Safe Self-Driving

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.

Automated summary

In this study, an end-to-end self-driving network with a sparse attention module that automatically attends to important regions of the input is proposed. The attention module specifically targets motion planning, leading to improved safety and interpretability of the planner.