Modelling operator's skill by machine learning
Ivan Bratko · Information Technology Interfaces · 2000
Controlling complex dynamic systems requires skills that operators often cannot completely describe, but can demonstrate. This paper describes some research into the transfer of human control skill into an automatic controller. Controllers are generated from examples of control traces. This process can be aided by techniques of Machine Learning (ML), and is also called cloning. The paper gives a review of ML-based approaches to behavioural cloning, representative experiments, and an assessment of the results. Some recent work is discussed, including the extraction of the operator's subconscious sub-goals and the use of qualitative representations. It is argued that the key to success is a suitable representation and decomposition of the machine learning problem involved.