Near On-Policy Experience Sampling in Multi-Objective Reinforcement Learning.
Shang WangMathieu ReymondAthirai A. IrissappaneDiederik M. RoijersPublished in: AAMAS (2022)
Keyphrases
- multi objective
- reinforcement learning
- optimal policy
- policy search
- approximate policy iteration
- evolutionary algorithm
- action selection
- optimization algorithm
- multi objective optimization
- markov decision process
- partially observable
- multiple objectives
- policy iteration
- reinforcement learning problems
- control policy
- objective function
- partially observable environments
- action space
- policy evaluation
- function approximation
- reinforcement learning algorithms
- function approximators
- nsga ii
- approximate dynamic programming
- pareto optimal
- genetic algorithm
- state and action spaces
- particle swarm optimization
- reward function
- state action
- markov decision problems
- multi objective optimization problems
- state space
- random sampling
- optimal control
- monte carlo
- infinite horizon
- dynamic programming
- actor critic
- policy gradient
- dynamical systems
- search space
- markov decision processes
- partially observable markov decision processes
- continuous state spaces
- partially observable domains
- multi agent
- learning process
- agent learns
- long run
- optimization problems
- state dependent
- sample size
- average reward
- continuous state
- utility function
- rl algorithms
- conflicting objectives