Contextual Pre-planning on Reward Machine Abstractions for Enhanced Transfer in Deep Reinforcement Learning.
Guy AzranMohamad H. DaneshStefano V. AlbrechtSarah KerenPublished in: AAAI (2024)
Keyphrases
- reinforcement learning
- transfer learning
- action selection
- macro actions
- state space
- model free
- function approximation
- complex domains
- reinforcement learning algorithms
- eligibility traces
- contextual information
- reward shaping
- partially observable
- knowledge transfer
- stochastic domains
- heuristic search
- learning agent
- markov decision processes
- reinforcement learning problems
- multi agent
- markov decision problems
- temporal abstractions
- batch processing
- deterministic domains
- planning problems
- partially observable markov decision processes
- learning algorithm
- total reward
- high level
- machine learning
- mixed integer programming model
- blocks world
- function approximators
- action space
- ai planning
- reward function
- flowshop
- learning tasks
- optimal policy
- dynamic programming
- learning process