BriefGPT.xyz
Oct, 2018
可微分模型预测控制（MPC）用于端到端规划与控制
Differentiable MPC for End-to-end Planning and Control
HTML
PDF
Brandon Amos, Ivan Dario Jimenez Rodriguez, Jacob Sacks, Byron Boots, J. Zico Kolter
TL;DR
该研究提出了一种利用模型预测控制（MPC）作为可微政策类来学习连续状态和行动空间中的强化学习的基础，通过使用控制器固定点处的凸逼近的KKT条件区分MPC，从而能够学习控制器的成本和动力学，旨在提高数据效率并优于传统系统识别。
Abstract
We present foundations for using
model predictive control
(MPC) as a differentiable policy class for
reinforcement learning
in continuous state and action spaces. This provides one way of leveraging and combining
→