Model–free least–squares policy iterationModel–free least–squares policy iteration Πλήρης Δημοσίευση σε Συνέδριο Conference Full Paper 2015-11-142001enWe propose a new approach to reinforcement learning which combines least squares function approximation with policy iteration. Our method is model-free and completely o policy. We are motivated by the least squares temporal dierence learning algorithm (LSTD), which is known for its ecient use of sample experiences compared to pure temporal dierence algorithms. LSTD is ideal for prediction problems, however it heretofore has not had a straightforward application to control problems. Moreover, approximations learned by LSTD are strongly in uenced by the visitation distribution over states. Our new algorithm, Least-Squares Policy Iteration (LSPI) addresses these issues. The result is an o-policy method which can use (or reuse) data collected from any source. We test LSPI on several problems, including a bicycle simulator in which it learns to guide the bicycle to a goal eciently by merely observing a relatively small number of completely random trials. http://creativecommons.org/licenses/by/4.0/1547–1554Neural Information Processing SystemsProceedings of NIPS*2001: Neural Information Processing Systems, Vancouver, BC, Canada Lagoudakis Michael Λαγουδακης Μιχαηλ Parr, R. Artificial Intelligence