Control of an Autonomous Lawnmower

A presentation on the final work on my undergraduate senior year thesis

1. CONTROL OF AN AUTONOMOUS LAWNMOWER Project Advisor: Dr. Abubakr Muhammad Group Members: • Haseeb Tariq 2014-10-0040 • Zia ul Azam 2014-10-0106 • Salman Nazir 2014-10-0110 EE- Senior Year Project
2. Agenda  Project overview  Last semester review  Improvements to the robot  Higher level control with ROS  Obstacle avoidance  Simulations and results  Issues  Future extensions
3. Project Aims  Equip a given mechanical lawnmower setup with the motor drives and other required circuitry  Establish a lower level of control for the basic locomotion of the lawnmower  Integrate the setup with a higher level of control for decision making and path planning  Make the robot autonomous to a certain degree by implementing an obstacle avoidance algorithm
4. Review of Last Semester
5. Existing Setup  Constraints to work with:  Left wheel independent, right wheel and cutting cylinder connected together  1horsepower motor for each wheel  Left side powered by 12V and right side by 24V sealed lead acid batteries
6. Key Features  H-bridge Motor Drives  High Current 20 Amperes  Power Mosfets, Gate Drivers and Power Diodes  Arduino Mega 2560  Lower level control for basic directional movement  PWM channels and PID control
7. Key Features  Optical Encoders  Measure wheel rotation and speed  Infrared sensors reflect off the wheel pattern  Wireless Remote Control  Interfaced with Arduino to test for basic directional control
8. Improvements made this Semester
9. H-bridges  Various improvements:  Isolation of control circuitry from High Current circuitry  Parallelized MOSFETS to increase current handling capability of H-bridges  Fuses for protection
10. Encoders  Issues with previous encoders:  Dirt and grass stuck to tires made the encoders prone to false clicks in readings  Solution:  Incremental shaft encoders (1000 pulses per revolution)  Stable and more precise
11. Mechanical Structure  Problem:  Two out of three batteries were on top tray  Short base and high center of gravity  Solution:  Expansion of lower tray to house the batteries  Extension of the rear rollers to increase the base area
12. Circuit Box  Initially, the circuitry was on the top tray and prone to short circuit due to vibrations and movement  Now, each component is properly housed and screwed in the box  PC fans are used for cooling purposes
13. Laser Mount Laser Scanner mounted to the base frame by a joint that offers 2 DOF:  Prismatic  Revolute Provides flexibility for later recalibrations or different requirements
14. Miscellaneous Improvements  Smoothing out of wheels and rotation blades  Changed the bearings of the rotating cutter cage  Cleaned and greased the inner rack and pinion lining of the wheels  Result  Wheels and cutter cage rotate more easily and draw less current
15. How to turn the Robot Precisely ?
16. Digital Compass  CMPS 10 Magnetometer was used for sensing the orientation of robot with respect to Magnetic North  It was necessary for ensuring the precise turning of Robot based on the angle communicated by Motion Planner in ROS to the ROS node running on Arduino
17. Angle Control Strategy Proportional Control Reference Heading Motion Planner Lawnmower Drive Magnetometer Required Change in Angle Current Orientation Magnetic Flux Density Error Current Heading
18. Issues with Magnetic Sensor  Sensor noise was encountered due to:  Magnetic shielding caused by metallic body  Magnetic interference caused by current carrying power wires  Solution: The magnetometer was vertically elevated from the base of Lawnmower
19. Robot Operating System
20. Robot Operating System  ROS is an open-source meta-OS used for designing Robotic Applications for High-level Control  It has Two Concept Levels:  Computation Graph Level  File System Level  Packages  Package Manifest, Message and Service Types
21. High Level Computational Graph ROS Serial Library Arduino Mega Laser Range Finder Motion Planner Laser Data Angle & Direction
22. Path Planning and Obstacle Avoidance
23. Range Sensing  Hokuyo Laser Range Finder  Coverage: 2D - 270˚ span  Range: 0.023 to 60 meters  Angle resolution: 0.25˚  USB interface  ROS Hokuyo node available to publish data on LaserScan topic
24. Open Path Algorithm  Objective:  To search the region scanned by the laser range finder for a direction that will allow a robot of specified width to move the furthest distance  Von Wahlde, Raymond. Wiedenman, Nathan. Brown, Wesley A. Viqueira, Cezarina. (2009). An Open-Path Obstacle Avoidance Algorithm Using Scanning Laser Range Data. Ft. Belvoir: Defense Technical Information Center
25. Open Path Algorithm  A laser scan returns an array which defines the environment  Alternate representation as an obstacle array Cartesian Coordinates Obstacle Array
26. Open Path Algorithm  Checks for all the paths available to the robot at each heading and then selects the one with the maximum area
27. Simulation in Gazebo
28. Issues Faced in Porting the Algorithm to the Actual Platform From Simulation to Reality
29. Limited Mobility  Problem:  Robot could not execute rotation at the minimum angle resolution of 0.25˚ as dictated by our algorithm  Solution:  A buffer of 5 degrees was allowed to be considered within range of the desired heading  The path width of the robot was extended to compensate for the loss in precision
30. Laser Scanner Imperfections  Problem:  Sensor noise due to resolution of laser beams on edges causing fluctuations in readings  Random spikes in sensor readings creating outliers  Solution:  A running median filter was applied to the raw laser data to remove the outliers and smooth out the readings
31. Practice Run Results
32. Further Work  Achieve tilt compensation of the compass to make it robust to uneven terrain of outdoor environments  Implement a wireless kill switch to remotely disable the robot to make it safer for outdoor testing  Configure the acceleration and deceleration of the robot for an outdoor environment
33. Future Extensions of the Project  Terrain classification algorithms can be used to distinguish between grass and non grass areas  A full fledged coverage and mapping algorithm like 2D slam could be implemented on top of the obstacle avoidance algorithm to provide full coverage of a lawn
34. Thank You for Listening Q&A