Learning HTN method preconditions and action models from partial observations

摘要: To apply hierarchical task network (HTN) planning to real-world problems, one needs encode the HTN schemata and action models beforehand. However, acquiring such domain knowledge is difficult time-consuming because definition involves a significant knowledge-engineering effort. A system that can learn automatically would save time allow be used in domains where knowledgeengineering effort not feasible. In this paper, we present formal framework algorithms acquire knowledge, by learning preconditions effects of actions methods. Our algorithm, HTN-learner, first builds constraints from given observed decomposition trees build method preconditions. It then solves these using weighted MAX-SAT solver. The solution converted Unlike prior work on learning, do depend complete or state information. We test algorithm several domains, show our HTN-learner both effective efficient.