Control of an Acrobot system using reinforcement learning with probabilistic policy search

Reinforcement learning with probabilistic policy search method is used in this paper for controlling an Acrobot system. Reinforcement learning with probabilistic policy search is a technique that is data-efficient and based on a model. Model bias is one of the main reasons for not using methods which are based on the model to learn from scratch. The model bias is not a severe problem in reinforcement learning with probabilistic policy search as it uses the Gaussian process which considers model uncertainty. Reinforcement learning with probabilistic policy search has the ability to give the best results even when very less data is available. The state of the art approximate inference is used for policy evaluation and for policy improvement. Policy gradients are calculated analytically.