Recently, effective algorithms to learn relational world models from experience in stochastic domains have been proposed. However, efficient planning with these models remains a major issue. We propose to convert learned noisy probabilistic relational rules into a structured dynamic Bayesian network representation. By predicting the effects of action sequences using approximate inference we are able to plan in complex worlds. We evaluate the practicality of our approach for online planning in a 3D simulated blocksworld with an articulated manipulator and realistic physics where we show that our method can solve problems where existing methods fail.
Discussion