Illustration of the hierarchy of decision-making processes. A destination is passed to a route planner that generates a route through the road network. A behavioral layer reasons about the environment and generates a motion specification to progress along the selected route. A motion planner then solves for a feasible motion accomplishing the specification. A feedback control adjusts actuation variables to correct errors in executing the reference path.
Questions:
- 根据什么来决定此时的行为?(超车?跟车?变道?)
- cost function?
- 决定超车后,如何规划一条可行的轨迹?
- 不同框架PNC模块的输入是什么?
- Apollo ?
- Waymo ?
- 什么时候会有行为层(behavioral layer)?
- Apollo中是否有行为层?
- 哪些模块可以用学习的方法(如:RL, DL)得到?
- PNC目前存在的难点是什么?
参考资料
- 2016 - MIT - A Survey of Motion Planning and ControlTechniques for Self-Driving Urban Vehicles
- 《Creating Autonomous Vehicle Systems》Chapter 07: Decision,Planning and Control [English] [Chinese] [Mindnote]
- 刘少山:无人驾驶的决策规划控制技术(雷锋网)