Model-Based Policy Search for Automatic Tuning of Multivariate PID Controllers

摘要

PID control architectures are widely used in industrial applications. Despitetheir low number of open parameters, tuning multiple, coupled PID controllerscan become tedious in practice. In this paper, we extend PILCO, a model-basedpolicy search framework, to automatically tune multivariate PID controllerspurely based on data observed on an otherwise unknown system. The system'sstate is extended appropriately to frame the PID policy as a static statefeedback policy. This renders PID tuning possible as the solution of a finitehorizon optimal control problem without further a priori knowledge. Theframework is applied to the task of balancing an inverted pendulum on a sevendegree-of-freedom robotic arm, thereby demonstrating its capabilities of fastand data-efficient policy learning, even on complex real world problems.