Non-Linear Approach for the Unmanned, Compound Helicopter Control

Abstract

The paper presents part of the work conducted within the project "Automatic Control of a Compound Helicopter" led by Warsaw University of Technology and sponsored by the BOEING Company. The aim of the paper is to present a proposal for the realisation of a control system for a small-scale compound helicopter utilizing one category of machine learning - reinforcement learning. For a given purpose, the theoretical foundations of reinforcement learning are presented and the DDPG methods, as a proposed approach for solving the control issue, is described. The comparison between the traditional control system and the way reinforcement learning works demonstrates the applicability of the artificial intelligence subcategory to control issues. The presented control object, helicopter ARCHER, is developed and evaluated in FLIGHTLAB software. An important part of the paper is the suggested implementation of the reinforcement learning method in MATLAB / SIMULINK software. Further plans for the development and testing of the control system are also revealed.