AA 274

Principles of Robotic Autonomy

The Robot Operating System (ROS)
Writing Software for Robotics
l Robotics requires very
complex software
l The software you will

deal with in AA274A

has way more moving
parts than what you’ve
dealt with in most
other classes…

9/23/21 AA 274A | Lecture 2 2

Writing Software for Robotics
l We deal with the complexity through modularity
l We enable modularity by following the right design pattern: “a
general, reusable solution to a commonly occurring problem within a given
context in software design” – Wikipedia

9/23/21 AA 274A | Lecture 2 3

The Pub/Sub Design Pattern
l We divide our software into individual components
l We define “topics” (think chat rooms) where components can
broadcast information to anyone listening
l Each component can:
l Publish: send messages to a topic regardless of whether someone is listening or not
l Subscribe: receive messages on a topic if anyone is sending them regardless of who

9/23/21 AA 274A | Lecture 2 4

The Pub/Sub Design Pattern

Note: there are countless ways to IMPLEMENT pub/sub!

9/23/21 AA 274A | Lecture 2 5

The Pub/Sub Design Pattern

You already use

Note: there are countless ways to IMPLEMENT pub/sub!
Pub/Sub every day!
Where???
9/23/21 AA 274A | Lecture 2 6
Alternatives to Pub/Sub
l Request/Reply (RPC)
l Push/Pull
l Data binding (e.g. shared data members)

l Observers

9/23/21 AA 274A | Lecture 2 7

What is ROS?
Depending on who you are talking to…
l An implementation of pub/sub geared towards robotic applications
and that is network-aware
l Lots of open-source software shared by the community:
n SLAM (gmapping, amcl)
n Vision (OpenCV, PCL, OpenNI)
n Arm Navigation (MoveIt)
n Simulation (Gazebo)

9/23/21 AA 274A | Lecture 2 8

Are there “Alternatives” to ROS?
l LCM
l Drake
l Player

l YARP
l Orocos
l MRPT

l And many others!

9/23/21 AA 274A | Lecture 2 9

Why is ROS popular in industry?
l Not reinventing the wheel is generally good
l Robotics is hard! It’s great to offload some of the work to smart
people
l ROS is now 12 years
old and still going
strong

9/23/21 AA 274A | Lecture 2 10

Why are we using ROS in AA274?
l The closest thing we have to an “industry standard”
l It’s an insurance policy for you (stability, online teaching resources)

9/23/21 AA 274A | Lecture 2 11

Why not ROS 2?
• A major overhaul of ROS
• Ecosystem/documentation still not quite as
complete
• Fundamental design pattern (if not
implementation mechanics) still the same
• Keep an eye on it!

http://design.ros2.org/articles/why_ros2.html

9/23/21 AA 274A | Lecture 2 12

ROS – Robot Operating System
• 2007-Today
l Stanford AI Robot (STAIR)
l Willow Garage founded by Scott Hassan (eGroups, Google,
Stanford Digital Libraries)
l Willow awards 11 $400k PR2 robots to Universities
l OSRF (Open Source Robotics Foundation) created to maintain ROS
and Gazebo
l ROS is everywhere!

9/23/21 AA 274A | Lecture 2 13

ROS Integrates Existing Projects
l OpenCV (computer vision)
l Stage, Gazebo (simulation)
l OpenSLAM (navigation)
l Orocos KDL (arm navigation)
l Many ROS “wrappers” to existing software

9/23/21 AA 274A | Lecture 2 14

The Main Software Components
1) Master
2) Nodes

l Nodes talk to each other over topics (think chat rooms). Master
coordinates the whole thing
l Message types: abstraction away from specific hardware

l Camera image
l Laser scan data
l Motion control

9/23/21 AA 274A | Lecture 2 15

ROS Node
l A process (typically Python or C++) that runs some computation
l The “fundamental” building block
l Can act as a subscriber, publisher or both

l Nodes talk to each other over “topics”

l Run them using rosrun <package> <node>
l Initialize using rospy.init_node()

Note: nodelets are different. They are not individual processes, they share memory

9/23/21 AA 274A | Lecture 2 16

Node Examples
Sensors and actuators are
wrapped in self-contained,
reusable software containers
called “nodes”

9/23/21 AA 274A | Lecture 2 17

Node Examples
Higher level operations also
become nodes in the ROS
computational architecture

9/23/21 AA 274A | Lecture 2 18

More Concrete Node Examples
l LiDAR node publishes laser scan arrays
l Camera node publishes RGB images (+depth if RGBD) and camera info
(resolution, distortion coefficients)
l Mobile robot controller publishes odometry values (e.g. x-y
coordinates and velocities, +z for UAVs or underwater vehicles)
l Navigation node subscribes to LiDAR and odometry messages,

publishes motion control messages

9/23/21 AA 274A | Lecture 2 19

ROS Master
l A process that is in charge of coordinating nodes, publishers and
subscribers
l Also provides a global parameter server

l Exactly one of them running at any time

l Messages do NOT go through Master (i.e. peer-to-peer)
l Nodes will not be able to find each other without Master

9/23/21 AA 274A | Lecture 2 20

Sending Messages
l pub = rospy.Publisher()
l msg = ...
l pub.publish(msg)

9/23/21 AA 274A | Lecture 2 21

ROS Node - Publisher

9/23/21 AA 274A | Lecture 2 22

Monitoring Messages
l You can check if you are sending messages using the rostopic
command line tool:

l rostopic list – lists all the active topics

l rostopic echo <topic> – prints messages received on <topic>
l rostopic hz <topic> – measures topic publishing rate

9/23/21 AA 274A | Lecture 2 23

Receiving Messages
l rospy.Subscriber("chatter", String, callback)
l def callback(msg): …

(in C++ need to call spinOnce(), not in Python)

9/23/21 AA 274A | Lecture 2 24

ROS Node - Subscriber

9/23/21 AA 274A | Lecture 2 25

ROS Launch Files
l Simple XML files that allow you to
n Launch multiple nodes at once
n Set parameters for those nodes
n Start Master

l roslaunch <package> <file>.launch

9/23/21 AA 274A | Lecture 2 26

ROS Launch File Example
<launch>
<!-- Start the talker node -->
<node name="talker" pkg="aa274" type="talker.py" output="screen">
<param name="rate" value="5"/>
</node>
</launch>

9/23/21 AA 274A | Lecture 2 27

A Case Study
• Edge detection in camera images

Node 1 – Camera Driver Node 3 – image_view

Subscribes to: Nothing Subscribes to: Camera images
Publishes: Camera images Publishes: Nothing

Node 2 – Edge Detection Node 4 – image_view

Subscribes to: Camera images Subscribes to: Image with edges
Publishes: Image with edges Publishes: Nothing

9/23/21 AA 274A | Lecture 2 28

A Case Study
l Edge detection in camera image
l rqt_graph

9/23/21 AA 274A | Lecture 2 29

ROS Launch File for Edge Detection
<launch>
<arg name="video_device" default="/dev/video0" />

<include file="$(find aa274)/launch/usbcam_driver.launch">

<arg name="video_device" value="$(arg video_device)" />
</include>

<node name="image_view_1" pkg="image_view" type="image_view">

<remap from="image" to="/camera/image_color" />
<param name="autosize" value="true"/>
</node>

<node name="image_view_2" pkg="image_view" type="image_view">

<remap from="image" to="/edge_detection/image" />
<param name="autosize" value="true" />
</node>

<node name="edge_detection" pkg="opencv_apps" type="edge_detection">

<remap from="image" to="/camera/image_color" />
<param name="debug_view" value="false" />
</node>
</launch>
9/23/21 AA 274A | Lecture 2 30
Developing with ROS
l Catkin workspace: a directory that contains all your ROS development
l It sets the right environment variables
l It knows how to compile your nodes (using cmake which in turn uses

a compiler)

The commands you need to know:

l mkdir -p ~/catkin_ws/src
l cd ~/catkin_ws
l catkin_make

9/23/21 AA 274A | Lecture 2 31

ROS Packages
l The basic organization structure for your nodes
l Usually corresponds to a “functionality” (e.g. a SLAM package)
l Can contain code for multiple nodes

l Directory structure:

The command you need to know:

catkin_create_pkg <name> roscpp rospy std_msgs

9/23/21 AA 274A | Lecture 2 32

Debugging
l rospy.loginfo()
l rqt_console
l rosbag record <topic>
l rosbag play file.bag

l pdb – Python Debugger

l import pdb

l pdb.set_trace()

9/23/21 AA 274A | Lecture 2 33

Creating Custom Messages
l Write message definitions (.msg) that are language agnostic
l ROS generates the right files so that roscpp and rospy can use your

message
l rosmsg show student

[aa274/Student]:
string name_first
string name_last
uint8 age
uint32 grade

9/23/21 AA 274A | Lecture 2 34

ROS Services
l A different way for nodes to pass messages to each other
l Request/Response scheme (not Pub/Sub!)
l Examples:

l Turn a light or LED on or off

l Assign a name to a face and retrain face recognizer
l Spawn a new model in the Gazebo simulator

9/23/21 AA 274A | Lecture 2 35

The Parameter Server
l Parameters are stored under namespaces; e.g.
n /move_base/local_costmap/height
n /usb_cam/framerate
n /gazebo/time_step
l Setting and getting parameters:
l rosparam set param_name param_value
l param_value = rospy.get_param("param_name")
l NOTE: Setting a parameter does not affect a running node!

9/23/21 AA 274A | Lecture 2 36

Dynamic Reconfigure
l Some nodes provide dynamically changeable parameters
n rosrun rqt_reconfigure rqt_reconfigure

9/23/21 AA 274A | Lecture 2 37

URDF
l Universal Robot Description Format
l An XML file that describes the kinematic chain of your robot

9/23/21 AA 274A | Lecture 2 38

Gazebo
l Same code that will run in production
l Physics is mostly accurate

9/23/21 AA 274A | Lecture 2 39

Some more libraries you will hear about...
l TF: coordinate frame transform library
l Actionlib: processes with goals and feedback
l dynamic_reconfigure: making nodes configurable on the fly

9/23/21 AA 274A | Lecture 2 40

Getting help
l ROS wiki (http://wiki.ros.org/)
l Github
l Stack Overflow

l The Construct / Robot Ignite Academy

l Google :)

9/23/21 AA 274A | Lecture 2 41

Next time
• Motion control

9/23/21 AA 274A | Lecture 2 42