Offline Reinforcement Learning with Mixture of Deterministic Policies.
Takayuki OsaAkinobu HayashiPranav DeoNaoki MorihiraTakahide YoshiikePublished in: Trans. Mach. Learn. Res. (2023)
Keyphrases
- reinforcement learning
- optimal policy
- policy search
- markov decision process
- control policies
- deterministic domains
- markov decision processes
- state space
- reward function
- stationary policies
- mixture model
- hierarchical reinforcement learning
- reinforcement learning agents
- function approximation
- total reward
- control policy
- fitted q iteration
- long run
- reinforcement learning algorithms
- policy gradient methods
- partially observable markov decision processes
- markov decision problems
- temporal difference
- model free
- dynamic programming
- finite state
- learning algorithm
- real time
- action selection
- reinforcement learning methods
- policy iteration
- robotic control
- decentralized control
- function approximators
- macro actions
- initially unknown
- learning process
- partially observable
- probabilistic model
- expectation maximization
- gaussian mixture model
- state abstraction
- decision problems
- black box
- gaussian distribution