(一)paper传送门
Guided Policy Search Model-based Reinforcement Learning for Urban Autonomous Driving
(二)背景知识
强化学习(RL)---------------------------《Reinforcement Learning:An Introduction》第二版
Gaussian Mixed Model(GMM) --------- https://www.jianshu.com/p/9708bb0676b2
Dual Gradient Descent(DGD)-----------https://www.jianshu.com/p/e3e672311442
KL-divergence(KL散度)-------------- https://blog.csdn.net/ningyanggege/article/details/87906786
Carla(模拟器)--------------------------------https://carla.readthedocs.io/en/latest/
(三)摘要
本文提出了一种基于模型的强化学习算法:Guided Policy Search (GPS)来学习在复杂、密集的城市环境中自动驾驶。总体思想是:采用基于模型的思想和使用GMM首先对系统动态进行近似,然后用双梯度下降法(DGD)来优化约束的策略优化问题(受制于轨迹变化幅度的限制(