Reinforcement Learning for Classical Planning: Viewing Heuristics as Dense Reward Generators.
Clement GehringMasataro AsaiRohan ChitnisTom SilverLeslie Pack KaelblingShirin SohrabiMichael KatzPublished in: ICAPS (2022)
Keyphrases
- classical planning
- reinforcement learning
- state space
- heuristic search
- partially observable
- planning problems
- heuristic function
- domain independent
- larger problems
- reward function
- deterministic domains
- markov decision processes
- temporally extended
- reinforcement learning algorithms
- planning graph
- optimal policy
- ai planning
- temporal planning
- initial state
- learning algorithm
- machine learning
- markov decision process
- belief space
- dynamic programming
- multi agent
- markov chain
- orders of magnitude
- belief state
- partial observability
- forward search
- state variables
- general purpose
- learning agent
- web services
- optimal control