Reinforcement Learning for Classical Planning: Viewing Heuristics as Dense Reward Generators.
Clement GehringMasataro AsaiRohan ChitnisTom SilverLeslie Pack KaelblingShirin SohrabiMichael KatzPublished in: CoRR (2021)
Keyphrases
- reinforcement learning
- classical planning
- state space
- partially observable
- heuristic search
- planning problems
- domain independent
- larger problems
- deterministic domains
- heuristic function
- planning graph
- markov decision processes
- reward function
- temporally extended
- reinforcement learning algorithms
- optimal policy
- planning domains
- initial state
- learning agent
- belief space
- forward search
- state variables
- machine learning
- dynamic programming
- orders of magnitude
- temporal planning
- markov chain
- partial observability
- dynamical systems
- learning algorithm
- average reward
- optimal control
- sufficient conditions
- markov decision process
- search algorithm
- markov decision problems
- decision problems
- search strategies
- multi agent
- web services