Abstract:
We consider the policy search approach to reinforcement learning. We show that if a ``baseline distribution'' is given (indicating roughly how often we expect a good policy to visit each state), then we can derive a policy search algorithm that terminates in a finite number of steps, and for which we can provide non-trivial performance guarantees. We also demonstrate this algorithm on several grid-world POMDPs, a planar biped walking robot, and a double-pole balancing problem.
copied = false, 2000);
">
@conference{Bagnell-2003-8823,
author = {J. Andrew (Drew) Bagnell And Sham Kakade And Andrew Ng And Jeff Schneider},
title = {Policy Search by Dynamic Programming},
booktitle = {Proceedings of (NeurIPS) Neural Information Processing Systems},
year = {2003},
month = {December},
pages = {831 - 838},
publisher = {MIT Press},
keywords = {Reinforcement Learning, Control, Policy Search, POMDP, partial observability, dynamic programming},
}
author = {J. Andrew (Drew) Bagnell And Sham Kakade And Andrew Ng And Jeff Schneider},
title = {Policy Search by Dynamic Programming},
booktitle = {Proceedings of (NeurIPS) Neural Information Processing Systems},
year = {2003},
month = {December},
pages = {831 - 838},
publisher = {MIT Press},
keywords = {Reinforcement Learning, Control, Policy Search, POMDP, partial observability, dynamic programming},
}