Learning by observation and practice: An incremental approach for planning operator acquisition

From Cohen Courses
Jump to navigationJump to search

Citation

   author = {Xuemei Wang},
   title = {Learning by Observation and Practice: An Incremental Approach for Planning Operator Acquisition},
   booktitle = {In Proceedings of the 12th International Conference on Machine Learning},
   year = {1995},
   pages = {549--557},
   publisher = {Morgan Kaufmann}


Online version

Paper: [1]


Abstract from the paper

This paper describes an approach to automatically learn planning operators by observing expert solution traces and to further refine the operators through practice in a learning-by-doing paradigm. This approach uses the knowledge naturally observable when experts solve problems, without need of explicit instruction or interrogation. The inputs to our learning system are: the description language for the domain, experts’ problem solving traces, and practice problems to allow learning-by-doing operator refinement. Given these inputs, our system automatically acquires the preconditions and effects (including conditional effects and preconditions) of the operators. We present empirical results to demonstrate the validity of our approach in the process planning domain. These results show that the system learns operators in this domain well enough to solve problems as effectively as human-expert coded operators. Our approach differs from knowledge acquisition tools in that it does not require a considerable amount of direct interactions with domain experts. It differs from otherwork on automaticallylearning operators in that it does not require initial approximate planning operators or strong background knowledge.

Summary

Writing planning domains in PDDL language needs knowledge of an expert person who has enought information about both domain and PDDL language. For many domains (e.g. solvnig math problems) there are many examles and traces available which can be used for programs to learn planning domain. This paper has developed a method to extract definition of a domain in PDDL by learning from a set of examples. They have assumed that all these examples are provided in a formal language. This assumption makes their work limited to specific applications.

The main advantage of their tecnique is that they don't need any initial state or initial definitions for planning domains. The input to their learning system is:

  • Description language for the domain: this includes types of objects and predicates which are used.
  • observation of an expert person: a sequence of actions which are executed by an expert.
  • a set of practice problems so that the learning system can check the trained domain on practice problems.

In the experimental results they have shown that the number of operators and average number of planning nodes (while solvnig practice problems) decrease when we provide more practice problems.