Comparing Observation and Action Representations for Deep Reinforcement Learning in MicroRTS.
Shengyi HuangSantiago OntañónPublished in: CoRR (2019)
Keyphrases
- reinforcement learning
- action selection
- partially observable domains
- action space
- reward shaping
- multi agent
- state space
- extensive form games
- function approximation
- state action
- function approximators
- agent learns
- temporal difference
- reinforcement learning algorithms
- model free
- dynamic programming
- sensory inputs
- transition model
- unsupervised feature learning
- state and action spaces
- fitted q iteration
- deep learning
- deep belief networks
- action sequences
- machine learning
- symbolic representation
- dynamical systems
- higher level
- learning algorithm