Learning States Representations in POMDPLearning States Representations in POMDPContardo, Gabriella and Denoyer, Ludovic and Artières, Thierry and Gallinari, Patrick2013
Paper summaryopenreviewThe authors present a model that learns representations of sequential inputs on random trajectories through the state space, then feed those into a reinforcement learner, to deal with partially observable environments. They apply this to a POMDP mountain car problem, where the velocity of the car is not visible but has to be inferred from successive observations.
Learning States Representations in POMDP
Contardo, Gabriella
and
Denoyer, Ludovic
and
Artières, Thierry
and
Gallinari, Patrick
arXiv e-Print archive - 2013 via Local Bibsonomy
Keywords:
dblp
The authors present a model that learns representations of sequential inputs on random trajectories through the state space, then feed those into a reinforcement learner, to deal with partially observable environments. They apply this to a POMDP mountain car problem, where the velocity of the car is not visible but has to be inferred from successive observations.