CONTROL TECHNIQUES FOR MOBILE ROBOTS

Autonomous mobile robots are one of the most challenging robotic platforms to develop and deploy in structured and unstructured environments like warehouses and indoor/outdoor environments respectively. Perception modules play an important role in mobile robot navigation and path planning with obstacle avoidance. Mobile robot hardware is a sophisticated system comprising of actuators, sensors, computate power and lots of power transmission and motion mechanisms.

Autonomous navigation gets trickier due to different models of the robot. Conventional wheeled robots and modern bio-inspired robots are the most common mobile robots used in the industry and the academia and the different motion models accentuate the problems in effecting desired behavior on them. Mobile robot kinematics plays an important role in determing the type of control technique to be used.

It is essentially the different control techniques that utilize the robot kinematics and dynamics models to define laws that draw defined robot behavior. Different controllers used in mobile robot control are resposible for execution and successful completion of motion/trajectory defined by planners. Controllers used are usually complex closed-loop feedback systems with the robot perception.

The scope of this article is only limited to wheeled mobile robot control techniques, mostly for a differential drive or ackerman drive and any mention to mobile robot should essentially mean a wheeled mobile robot.

INTRODUCTION

Controllers deployed in mobile robots are usually only meant to follow the path specified by the user. They are essentially position controlled in the operational space. The control inputs, u to robot vary with the type of robot model. Examples are:

• Differential Drive (Powered rear wheels and free front castor wheel) - Inputs are the speeds for the two wheels, u = [vleft_wheel, vright_wheel]
• Ackerman Drive (Powered steering and powered coupled rear wheels) - Inputs are the speed for the rear wheels and the steering angle for the heading, u = [vrear_wheels, thetasteering]
• Mecannum Wheel Drive (All individually powered 3 or more wheels) - Inputs are the speed values for all the wheels. The control changes due to changed kinematic model from different mounting arrangments, u = [vwheel_1, vwheel_2, ..... vwheel_n]

CONTROL TECHNIQUES

Mobile robot control has two components, first a robot model that implements actuator instructions corresponding to the desired real-world motion and second, the closed-loop feedback mechanisms that provide observations of the actual situation.

The high-level trajectory defined by the motion planners are often partially or completely infeasible. Thus, execution of desired robot motion is also dependent on if the controller is able to execute each instruction. The ability

Control mechanisms for robots can be categorized into two broad categories - the high-level/robot-level control and the low-level/actuator-level control. Both levels of control are extremely important to realize the desired behavior of the robot.

The most common mobile robot control algorithms are:

• PID Controller
• Model Predictive Controller
• Pure Pursuit Controller
• Fuzzy Logic Controller
• Linear Quadratic Regulator

All these control techniques generate high-level instructions for mobile robot motion but it is the low level actuators that cause motion by implementing motion using the actual robot state, the environment around and the robot kinematic/dynamic model.

Note: A trajectory is a path (collection of points) with instructions for velocity, acceleration and

LINEAR QUADRATIC CONTROLLER

The LQR controller is a .

The high-level trajectory defined by the motion planners are often partially or completely infeasible. Thus, execution of desired robot motion is also dependent on if the controller is able to execute each instruction. The ability

PURE PURSUIT CONTROL

Pure Pursuit control is a tracking algorithm that generates a curvature for motion, from the current state to the goal state. .

The high-level trajectory defined by the motion planners are often partially or completely infeasible. Thus, execution of desired robot motion is also dependent on if the controller is able to execute each instruction. The ability