Adv-controls

Pre-requisites

MATLAB 2019Rb

Overview

This paper presents the work done on designingand implementing a nonlinear model predictive control (MPC)based controller on a self-balancing two wheeled mobile robotfor a dynamic environment. The inverted pendulum model usedin this study is naturally an unstable system when subjected tonatural disturbances due to environmental loading conditions.This model has been studied extensively in the past for PDcontrollers but MPC has been relatively left unexplored. Inthis paper we propose various controllers, PD controller,MPC controller as feedback and MPC reference trackingfeed forward controller with a inner PD controller, to drivethe balancing two wheeled robot in a dynamic environment.Simulation results demonstrate the feasibility and efficiency ofour proposed design and experimental results on the hardwarevalidate the these results. A follower robot and obstacle detectionwas added in the final demonstration to show multi roboticsinteraction.The PD controller with live camera feedback wasable to demonstatet robust line tracking, obstacle detection andmulti-robot interaction.The MPC based self balancing robotwas able to track a straight line with full state feedback.

Please read the report file to understand the project better

Instructions to run the files

1. nonlinear_plant.mlx contain all the nonlinear plant parameters required to run nonlinear_plant.slx Simulink model. nonlinear_plant.slx Simulink model contain both linear and nonlinear plant simulink models with a controller in loop to check for stability and reference tracking. Please run nonlinear_plant.mlx before and then run nonlinear_plant.slx to see simulation of output
2. Run MPC_feedback_position.mlx first and then MPC_feedback_position.slx to visualise the simscape nonlinear model behaviour. In this model we have position, veocity, tilt and rate of tilt as the statespace and here we track position using MPC feedback module.
3. Run MPC_feedback_velocity_ref.mlx first and then MPC_feedback_velocity.slx to visualise the simscape nonlinear model behaviour. In this model we have veocity, tilt and rate of tilt as the statespace and here we track velocity using MPC feedback module.
4. Run MPC_feedforward_position.mlx first and then MPC_feedforward_position.slx to visualise the simscape nonlinear model behaviour. In this model we have position, veocity, tilt and rate of tilt as the statespace and here we track position using MPC feedforward reference tracking with a innerloop PD controller module.
5. Run MPC_feedforward_velocity.mlx first and then MPC_feedforward_velocity.slx to visualise the simscape nonlinear model behaviour. In this model we have veocity, tilt and rate of tilt as the statespace and here we track position using MPC feedforward reference tracking with a innerloop PD controller module.
6. Position Tracking Feedback_linearplant.mlx implements feedback controller with MPC statespace as position,velocity,tilt and rate of tilt and additional PD controller for heading angle. Here we update the MPC with linearised plant equations instead of nonlinear plant.Please run the Position Tracking Feedback_linearplant.mlx to implement and visualise the tracking.
7. Position Tracking Feedforward_linearplant.mlx implements feedback controller with feedforward reference tracking MPC statespace as position,velocity,tilt and rate of tilt with inner loop PID controller and an additional PD controller for heading angle. Here we update the MPC with linearised plant equations instead of nonlinear plant.Please run the Position Tracking Feedback_linearplant.mlx to implement and visualise the tracking.

Arduino

1. Manual_line_track.ino

(Please upload Manual_line_track_slave.ino to the slave Nano first) (Upload Manual_line_track.ino to the master Nano. Set up hotspot on a mobile phone. On laptop open Putty and connect to the Raspberry Pi over the hotspot. This will open up mini-com which the user can use to give manual commands.) There are two possible modes of operation of the robot.

1.1. Manual Mode

    When switched on for the first time, robot boots up in the manual mode. If in correction mode press 'M' or 'm' to switch to manual mode. 
    In this mode the robot moves as follows:
    1.1.a. Press 'W' or 'w' to move the robot forward.
    1.1.b. Press 'S' or 's' to move the robot in reverse.
    1.1.c. Press 'A' or 'a' to turn the robot left in its current position.
    1.1.d. Press 'D' or 'd' to turn the robot right in its current position.
    1.1.e. Press 'Q' or 'q' to balance the robot in its current position.

1.2. Correction Mode (fully autonomous)

    Press 'C' or 'c' to switch from manual mode to correction mode. 
    In this mode, the user looses manual control of the robot. The user can only press 'M' or 'm' to switch back to manual mode.
    To run the robot in this mode, user will have to define a path on the ground with a tape (Dark tape on a light ground). The ground should not have any other lines or             patterns which could interfere with the Pixy's line tracking algorithm.
    Once setup is done, the robot will automatically track the line using reading obtained by the Pixy.

2. Manual_line_track_slave.ino

Upload this to the slave Nano when running Manual_line_track.ino in the master. This will send the x coordinate of the vector that is being tracked by the Pixy to the           master. This code only comes into play when user is running the robot in Correction mode. 

3. MPC.ino

(Open MPC_robot_square_wave.mlx in MATLAB on a computer)
This code will send currently calculated states to MATLAB running on a computer. MATLAB will then send the reference velocities to the master. 
This mode is also fully autonomous where user cannot interact with the ino.

The MPC mode is setup to take a reference velocity from the MATLAB script and track it using inner PI controller for velocity. The balancing is achieved using a PD control for tilt. The MATLAB script reads the states from the serial device (Raspberry Pi) and is fed as the start-point of the MPC formulation. Based on this and the reference square wave velocity it has to track, the MPC is formulated such that it provides the reference velocity while ensuring constraints on control effort [-12,12]V and maximum tilt angle.

Conclusions

In conclusion the Tumbller was able to balance and followlines with a PD controller effectively. For the final demothe team was able to incorporate obstacle detection and afollower robot. In terms of MPC the robot was able to tracka line of fixed distance. Full robot MPC was not achieveddue to the communication delays encountered in parsing theincoming data stream. From the limited testing performed dueto restricted on campus access the robot had a wide range offunctionality. This aligned with the team’s goal of making ascaled version of a self balancing multi-agent robot systemthat is able to function in as an automated guided vehicle.

Future work

In retrospect, based on the communication delays observedin running the paring scripts on MATLAB, the controllerwould have been better implemented if run locally on theRaspberry Pi either through MATLAB or an embedded scriptderived from MATLAB Coder. By generating C and C++ codefor the Raspberry Pi platform along with libraries for the IMUand Tumbller source code, a standalone version of the programcould be generated.