Mapping Instructions and Visual Observations to Actions with Reinforcement Learning.
Dipendra Kumar MisraJohn LangfordYoav ArtziPublished in: EMNLP (2017)
Keyphrases
- reinforcement learning
- perceptual aliasing
- action selection
- partially observable
- partial observations
- state transitions
- state and action spaces
- action space
- multiagent reinforcement learning
- visual information
- function approximation
- state action
- partially observable domains
- reinforcement learning algorithms
- situation calculus
- sensory inputs
- visual features
- loop closure
- partial observability
- learning algorithm
- reward function
- decision theoretic
- action sequences
- human activities
- initially unknown
- state space
- dynamic programming
- multi agent
- visual perception
- multiple agents
- optimal control
- markov decision processes
- action recognition
- low level
- neural network