Process control and machine learning: rule-based incremental control

In this paper, we discuss a rule-based incremental control program which has been used for controlling a laser cutting robot and in simulation for driving a car on a track, for a car parking manoeuvre, or for parking a truck with one trailer. The core of the paper concerns a learning program, Candide, which learns to control a process without a priori knowledge about the process, by observing random initial evolutions of the process and acquiring a qualitative model. Monotonous or derivative relationships between inputs and outputs are recognized, and then a rule-based incremental controller Is deduced from this model