Rhoban's workspace for Kid Size league

Setup of the system

Recommended Operating System

The recommended operating system to run this version of the software is Ubuntu 18.04 Bionic Beaver using other OS might result on issues with some of the packages required.

Installing APT dependencies

First of all, you will need to install required packages:

sudo apt-get install gcc cmake git libtinyxml-dev libncurses5-dev \
    php php-cli php-xml libv4l-dev gnuplot-qt \
    python3-pip python3-empy python3-setuptools python3-nose chrpath ffmpeg libudev-dev \
    libsfml-dev libconsole-bridge-dev freeglut3-dev libx11-dev libxrandr-dev libfreetype6-dev \
	    libjsoncpp-dev libprotobuf-dev protobuf-compiler libgtest-dev libtclap-dev \
    qt5-default qtmultimedia5-dev libqt5webkit5 \
    libopencv-dev liburdfdom-dev

Installing FlyCapture dependency

To use BlackFly cameras from FLIR, you have to install their software. First clone this repository outside of the workspace folder:

git clone https://github.com/RhobanDeps/flycapture.git

And run the install script:

cd flycapture
sudo ./install_flycapture.sh

Maybe there will be issues with apt packages, in this case, run:

sudo apt --fix-broken install

And try again (you might need to repeat the last step 2 or 3 times)

Setting up your Github account with your public key

Since the wks manager handle dozens of repositories at once, it is much more convenient to use SSH keys. If you don't have a one, generate one using:

ssh-keygen -t rsa

Sign in on your GitHub account and go to Settings, and then "SSH and GPG keys". Click "New SSH key" and copy the content of .ssh/id_rsa.pub in the key field, choose any name you want and validate the new key.

Setting up the workspace (rhoban developer)

First, install wks:

pip install wks

And then run:

wks install rhoban/kid_size

This will install the upstream repositories. You can now build using:

wks build

Adding Rhoban binaries to your path

Binaries are built in build/bin.

Run this command to add all rhoban binaries to your $PATH:

echo export "PATH=\"\$PATH:$PWD/build/bin\"" >> ~/.bashrc

Don't forget to re-run the shell to have the change applied. Once you build the rhoban tools, you should be able to use them without specifying the full path.

Workspace commands

To pull all the repositories:

wks pull

To build:

wks build

To build (debug), manually create a build_debug directory and simply run cmake ../src, then edit the proper CMakeCache variables using ccmake . to set the build type to DEBUG.

To build just a specific package:

wks build [package]

Workspace dependencies

Have a look at wks documentation for more details about how dependencies are handled.

Rhoban basic commands

Compiling the robot program (KidSize)

Run the following to build the program:

wks build KidSize

Communicating with the robot

Our robots communicate with the 10.0.0.1 ip address, so you need to configure your computer to a compatible static address like 10.0.0.2

It is strongly recommended that you add your private key to the robot. By copying the content of .ssh/id_rsa.pub in the .ssh/authorized_keys inside the robot.

Deploying the program to the robot

This will deploy the program on the robot:

./deploy

Running the program on the robot

This will remotely run the program on the robot

./run

Connecting to the robot with rhio

rhio 10.0.0.1

Working on robot logs

All of this commands have to be issued while located in the env/fake folder.

Preparing a log environment

In order to work with a specific log, use:

./prepare.py <path_to_log>

This command also set the serial_number of the tracker of the robot if available in the metadata.json file.

Extracting ground truth from Vive

It is possible to extract ground truth based on multiple logs with HTC Vive tracker

ln -sf ../common/vive_roi_extractor.json vision_config.json ./extract_vive_patches.py ...

In this case, all the data will be placed in the folder vive_data. It is important not to rewind the video (using key 'p') or update it stationary (using key 'u') while you extracting log data, because it has a risk of duplicating data. If you want to investigate on vive issues

Compressing vive data

It is possible to collect them in compressed tar.gz files to send them on distant servers faster:

./compress_vive_data.sh

This command will use the data in the folder vive_data. The patches used for classification will be stored in file classification_data.tar.gz and the images with the position of the objects will be stored in attention_data.tar.gz.

Labelling videos

This procedure is still evolving yet and will be made public along with data soon.

Using PyBullet viewer

The code allowing to display the robot with pybullet can be cloned outside of workspace:

git clone [email protected]:rhoban/sigmaban_pybullet.git

It requires some dependencies

sudo apt install python3-pip
sudo pip3 install -U zmq pybullet numpy protobuf

In order to use it, you have to enable publishing of the model using RhIO: set model/publish=true

Then you can launch the viewer from the sigmaban_pybullet folder

python3 client.py

Using rhoban_monitoring tool

The rhoban monitoring tool can be installed as following:

./workspace install rhoban/qt_monitoring

In order to view a game in progress, move to the folder where you want to write the log and use the following command:

rhoban_monitoring -l

A folder with a name based on current date will be created. You can replay it by moving in the folder and running:

rhoban_monitoring -r

For custom execution and extended options (name of the robots, cameras, etc...), you can edit a json configuration file (examples available in the qt_monitoring repository) and launch the element as follows:

rhoban_monitoring -m manager.json

Synchronising robot clocks

Documentation: doc/synchronisation.md.

Training new DNN

The code used to train multi-class DNN classifiers is independent from workspace and can be found on:

• https://www.github.com/rhoban/tf_deep_vision_public for use
• https://www.github.com/rhoban/tf_deep_vision for rhoban developers