Never Worse, Mostly Better: Stable Policy Improvement in Deep Reinforcement Learning.
Pranav KhannaGuy TennenholtzNadav MerlisShie MannorChen TesslerPublished in: AAMAS (2023)
Keyphrases
- reinforcement learning
- optimal policy
- policy search
- markov decision process
- reinforcement learning problems
- function approximation
- reinforcement learning algorithms
- control policy
- state space
- action selection
- partially observable environments
- dynamic programming
- markov decision problems
- markov decision processes
- policy gradient
- reward function
- partially observable
- infinite horizon
- partially observable domains
- state action
- temporal difference learning
- reinforcement learning methods
- control policies
- state and action spaces
- continuous state spaces
- transition model
- function approximators
- state dependent
- model free
- learning problems
- action space
- policy iteration
- asymptotically optimal
- temporal difference
- policy evaluation
- multi agent reinforcement learning
- sufficient conditions
- least squares
- multi agent systems
- robotic control
- multi agent
- learning algorithm