Planning without a domain model

Automated Planning focuses on plan search. Traditionally, it aimed at domain-independent methods with handcrafted domain models. However, automated domain model acquisition, especially the action model acquisition is difficult. On the other hand, many problem specific search space pruning techniques were proposed. Therefore, we combine the automated domain model acquisition and problem specific search space pruning to generate plans for all instances of a problem. To this end, we use deep learning techniques to learn planning behaviors, which already considered the domain models and the problem constrains. The biggest challenge of this method is encoding planning knowledge as the inputs of a deep neural network. Experiments showed that we can get instance-oriented perfect classifiers. With these classifiers, we can plan without planning models. Leveraging on transfer learning for generalization abilities is the most important future work.