Control of a Nonholonomic Mobile Robot Using Neural Networks

Introduction

该文章主要探讨了一个名为control structure的技术细节，并展示了它如何能够实现将kinematic控制器与neural network computed-torque控制器结合在一起的能力。

Mobile robot navigation is categorized into three fundamental challenges: tracking a reference trajectory, following a path, and achieving point stabilization. Nonlinear feedback controllers have been designed to address each of these challenges. The central concept of these algorithms is to determine appropriate velocity control inputs that ensure the stability of the closed-loop system. This article addresses all three fundamental challenges in mobile robot navigation.

在现有解决方案中, 通常情况下, 为了计算车辆控制输入, 假设存在完美速度跟踪. 这种假设将导致三个问题:

• The ideal velocity tracking assumption is invalid in practice
• Disturbances are overlooked in this context
• A comprehensive understanding of the system dynamics is essential

这篇文章探讨的方法实际上是利用一个NN控制器来应对上述的问题。
下面将详细介绍本文所提出的控制方法——非线性kinematic-NN回避式跟踪控制器：
该方法系统阐述了一种严谨的路径跟踪控制策略，在车辆动力学特性基础上实现了从理论指令到实际控制输入的精确转换。
因此该控制器本质上可划分为两大核心模块：

1. The feedback velocity control inputs are designed to be used in the kinematic steering system, resulting in asymptotic stability of the position error.
2. An NN-based computed-torque controller is designed for mobile robots, aiming at achieving convergence of their velocities towards specified input velocities.

第一部分可以是任意的kinematic-based approach，
第二部分does not rely on any navigation challenges, as its purpose is to calculate the required torque inputs by approximating the cart's nonlinear dynamics.