Continuous Reinforcement Learning With Knowledge-Inspired Reward Shaping for Autonomous Cavity Filter Tuning

Abstract

Reinforcement Learning has achieved a great success in recent decades when applying to the fields such as finance, robotics, and multi-agent games. A variety of traditional manual tasks are facing upgrading, and reinforcement learning opens the door to a whole new world for improving these tasks. In this paper, we focus on the task called Cavity Filter Tuning, a traditionally manual work in communication industries which not only consumes time, but also highly depends on human knowledge. We present a framework based on Deep Deterministic Policy Gradient for automatically tuning cavity filters, and design appropriate reward functions inspired by human expertise in the tuning task. Simulation experiments are conducted to validate the applicability of our algorithm. Our proposed method is able to autonomously tune a detuned filter to meet the design specifications from any random starting positions.