This week you should be working on Task1. Here are some technical details to help you with the implementation.
As you know by know, the Turtlebot4 simulation is notoriously compute intensive, and requires a dedicated GPU in order to be able to run the simulation fast enough so that it does not break completely. For this purpose, we set up six laptops in the classroom.
Each team has a username for each laptop, so you can use whichever one you like. Set up a Git repository for your code, since it is NOT guaranteed that you will find the files on the laptop in the same state as you left them. If you do not have the password for your team, ask the assistant.
As know, ROS2 uses a DDS system in order to manage the resources in a ROS2 graph. This means that all the nodes, and topics, and services on the same network can (in theory) find each other. This can be problematic when there are multiple people independently running ROS2 nodes, that should not communicate with each other. To fix this, we can set a ROS_DOMAIN_ID variable, so that our nodes will communicate only nodes that are running under the same ROS_DOMAIN_ID. To do this, you need choose a <number> between 0 and 101, and add the following line to the .bashrc file of all computers that you wish to be able to see each other (as far as ROS2 is concerned):
export ROS_DOMAIN_ID=<number>
Alternatively, if you will to only run ROS2 on a single computer, you can set:
export ROS_LOCALHOST_ONLY=1
In the case when Ignition Gazebo fails to start and you see repeated prints of Requesting list of world names., setting the IGN_IP will probably help:
export IGN_IP=127.0.0.1
If you have an Nvidia GPU, but you are still having problems with the simulation, it is very likely that Gazebo does not use the dedicated driver for the GPU. The most problematic result is that the laser is not being simulated, and as a result a map is not being built so you do not see the map in RViz. There are a lot of possible causes for this, this troubleshooting guide solved those problems for some students:
https://github.com/gazebosim/docs/blob/master/fortress/troubleshooting.md
As part of Task1, your robot needs to say something when it approaches a detected face. To do this, you can simply record an audio file, and use a module like playsound (that you install with pip install playsound). There are also other modules for playing sounds: pydub, simpleaudio, or using the os library and playing the sound with your system player.
A more interesting approach, and one that will also prove useful in the future, is using a text-to-speech (TTS) generator. This is currently a very active research field, and you have many different options for TTS generators. There are simple ones, complex ones, there are those that run on-device, and those that run in the cloud, you can even try to train a deep model for imitating some voice. For our purposes, the quality of the generated voice does not matter so do as you wish. One useful guide that you can use is: https://picovoice.ai/blog/on-device-text-to-speech-in-python/ .
If you do not want to write your own transformations (e.q. Euler angles to Quaternions or vice versa) you can use the following libraries:
sudo apt install ros-humble-tf-transformations
sudo apt install ros-humble-turtle-tf2-py
pip3 install transforms3d
Then, you can do stuff like:
import tf_transformations
q = tf_transformations.quaternion_from_euler(r, p, y)
r, p, y = tf_transformations.euler_from_quaternion(quaternion)
Those of you that can only work on the simulation in the lab, make use of the ros2 bag command line tool. It is a tool for recording all or some messages published. For example, you can run the simulation and drive to robot around the polygon, while recording the messages published (like the images from the camera). Then you can copy the bag file to another computer, play it, and work on face detection and clustering. The tutorial for ros2 bag is here.
In the demo package this week, you have a node that sends a navigation goal to the robot by clicking on the map. This demo package illustrates some ideas, like how you can read the map from the topic and convert it to a numpy image, how you can convert from pixel coordinates to real world coordinates and more. Download, build, test, explore code.
The other script in this package, demonstrates how you can use the TF2 libraries to do transformations between frames. This node first creates a point that is 0.5m in front of the robot (in its header the frame_id is "base_link"). Then it looks up the transformation between the "base_link" and the "map" frames, and then applies the transformation to the point so it is transformed in the "map" frame. Then it creates a marker from the point and publishes it, so you can see it in Rviz on the /breadcrumbstopic.