Policy Gradient Converges to the Globally Optimal Policy for Nearly Linear-Quadratic Regulators.
Yinbin HanMeisam RazaviyaynRenyuan XuPublished in: CoRR (2023)
Keyphrases
- policy gradient
- optimal policy
- average reward
- optimal control
- reinforcement learning
- dynamic programming
- partially observable markov decision processes
- infinite horizon
- actor critic
- markov decision processes
- decision problems
- finite horizon
- state space
- finite state
- long run
- function approximation
- average cost
- policy iteration
- reinforcement learning algorithms
- optimal solution
- initial state
- markov decision process
- single agent
- sufficient conditions
- closed loop
- function approximators
- gradient method
- model free
- multi agent