Andrew did ground-breaking work in unifying control theory with machine learning. Using a behavioural cloning framework he used decision trees to learn the goals for a controller. Linear model trees were then used to learn the parameters for PID controllers to implement the control actions.