Potential-based reward shaping using state-space segmentation for efficiency in reinforcement learning.
Melis Ilayda BalHüseyin AydinCem IyigünFaruk PolatPublished in: Future Gener. Comput. Syst. (2024)
Keyphrases
- reward shaping
- reinforcement learning
- state space
- reinforcement learning algorithms
- complex domains
- markov decision problems
- markov decision processes
- function approximation
- optimal policy
- heuristic search
- learning algorithm
- partially observable
- dynamic programming
- markov chain
- multi agent
- dynamical systems
- transfer learning
- infinite horizon
- model free
- action selection
- belief state
- supervised learning
- search space
- agent learns