Stabilizing Off-Policy Reinforcement Learning with Conservative Policy Gradients.
Chen TesslerNadav MerlisShie MannorPublished in: CoRR (2019)
Keyphrases
- reinforcement learning
- optimal policy
- policy search
- action selection
- markov decision process
- policy gradient
- partially observable environments
- state space
- actor critic
- control policy
- markov decision problems
- markov decision processes
- policy evaluation
- approximate dynamic programming
- function approximators
- policy iteration
- action space
- control policies
- reinforcement learning problems
- function approximation
- reward function
- partially observable
- decision problems
- state dependent
- state action
- control problems
- dynamic programming
- temporal difference learning
- state and action spaces
- temporal difference
- reinforcement learning algorithms
- long run
- finite state
- evaluation function
- continuous state
- partially observable markov decision processes
- learning process
- neural network
- robotic control
- decision making
- agent receives
- continuous state spaces
- rl algorithms
- average reward
- model free
- infinite horizon
- transition model
- policy making
- approximate policy iteration
- markov chain
- policy gradient methods
- model free reinforcement learning