Preferred-Action-Optimized Diffusion Policies for Offline Reinforcement Learning.
Tianle ZhangJiayi GuanLin ZhaoYihang LiDongjiang LiZecui ZengLei SunYue ChenXuelong WeiLusong LiXiaodong HePublished in: CoRR (2024)
Keyphrases
- reinforcement learning
- optimal policy
- fitted q iteration
- control policies
- policy search
- action space
- action selection
- markov decision process
- markov decision processes
- discounted reward
- partially observable domains
- function approximation
- state action
- initial state
- state space
- reward function
- reinforcement learning agents
- reward shaping
- continuous state
- state and action spaces
- partially observable markov decision processes
- anisotropic diffusion
- dynamic programming
- control problems
- macro actions
- model free
- markov decision problems
- transition model
- action sets
- hierarchical reinforcement learning
- decision problems
- state abstraction
- long run
- reinforcement learning algorithms
- sensory inputs
- policy gradient methods
- infinite horizon
- expected reward
- real time
- diffusion process
- control policy
- agent learns
- temporal difference
- decentralized control
- multi agent reinforcement learning
- function approximators
- average reward