Analysis of Model-Free Reinforcement Learning Control Schemes on self-balancing Wheeled Extendible System

TL;DR

This paper explores the application of deep reinforcement learning algorithms, specifically Deep Deterministic Policy Gradient and Proximal Policy Optimization, to control a self-balancing extendable wheeled system, demonstrating improved adaptability over traditional methods.

Contribution

It introduces RL-based control schemes for a complex nonlinear system and compares their performance with Model Predictive Control, highlighting their effectiveness and self-tuning capabilities.

Findings

RL controllers outperform MPC in trajectory tracking accuracy

Deep RL models adapt better to system dynamics

Self-tuning parameters improve control stability

Abstract

Traditional linear control strategies have been extensively researched and utilized in many robotic and industrial applications and yet they do not respond to the total dynamics of the systems. To avoid tedious calculations for nonlinear control schemes like H-infinity control and predictive control, the application of Reinforcement Learning(RL) can provide alternative solutions. This article presents the implementation of RL control with Deep Deterministic Policy Gradient and Proximal Policy Optimization on a mobile self-balancing Extendable Wheeled Inverted Pendulum (E-WIP) system with provided state history to attain improved control. Such RL models make the task of finding satisfactory control schemes easier and responding to the dynamics effectively while self-tuning the parameters to provide better control. In this article, RL-based controllers are pitted against an MPC controller to evaluate the performance on the basis of state variables and trajectory errors of the E-WIP system while following a specific desired trajectory.