Reinforcement learning to train Ms. Pac-Man using higher-order action-relative inputs.

Luuk Bom Ruud Henken Marco A. Wiering

Published in: ADPRL (2013)

Keyphrases

pac man
higher order
reinforcement learning
game tree search
action selection
sensory inputs
partially observable domains
action space
markov random field
state action
reward shaping
optimal policy
pairwise
markov decision processes
state space
reward function
dynamic programming