Learning-Deep-Learning

Jointly Learnable Behavior and Trajectory Planning for Self-Driving Vehicles

June 2024

tl;dr: Learning BP and TP jointly for ML-based motion planner.

Overall impression

This paper describes a way to “learn” a mmotion planner. Yet in reality the learning happens only in the relative weights of the already hand-crafted cost functions.

The paper has a very nice introduction what is motion planning, and the two components behavior planning and trajectory planning. The paper gives a very good high level review of the motion planning problem without poluting the main text with arcane math formulae without too much information.

Key ideas

Technical details

Notes