Qualitative Reconstruction of Control Skill

Controlling complex dynamic systems requires skills that operators often cannot completely describe, but can demonstrate . This paper is concerned with the problem of transfer of human control skill into an automatic controller . The process of reconstructing a skill from an operator's behavioural traces by means of Machine Learning (ML) techniques is called behavioural cloning . This paper focusses on the questions of appropriate representation for behavioural cloning when the main goal of cloning is to provide insight into operators' subcognitive skills, and not only to induce a successful controller . Disadvantages of traditional, numerical representations are discussed, and benefits of qualitative representations are demonstrated in a case study .