Table of Contents

1. Simple Installation
2. Manual Installation
   • Normal
   • Conda
3. Running
4. Updating
5. Docker

Simple Installation

1 . To install ros2 and other dependencies.

git clone https://github.com/siw-engineering/biped_ros2 ~/biped_ros2
cd ~/biped_ros2
bash biped_ros2_install.sh

Note: Do not run step 2 if you are using a conda environment, see conda section

2 . Once install is finished, close the current terminal and open a new one

cd ~/biped_ros2
rosdashing
colcon build --symlink-install
# The --symlink-install is such that instead of creating new files in the install folder, it just points to the files in the source folder, i.e. no copied files, easier to iterate changes

Manual Installation

Normal

1. Install ros2 dashing following the instructions here

2. Install ros2 development tools following the section named Install development tools and ROS tools listed here. The commands have also been copy pasted here for reference (with cyclone DDS dependencies removed).

sudo apt update && sudo apt install -y \
  build-essential \
  cmake \
  git \
  python3-colcon-common-extensions \
  python3-pip \
  python-rosdep \
  python3-vcstool \
  wget

3. Install Opensplice DDS sudo apt install ros-dashing-rmw-opensplice-cpp

4. Install Gazebo sudo apt install gazebo9 libgazebo9-dev

5. Install other dependencies sudo apt install libyaml-cpp-dev ros-dashing-python-qt-binding ros-dashing-gazebo*

6. Download the relevant source code. The optional repositories are mainly for URDF development as well as the development of binaries of utility libraries.

git clone https://github.com/siw-engineering/biped_ros2 ~/biped_ros2
cd ~/biped_ros2
mkdir src
vcs import src < main.repos

(optional) vcs import src < optional.repos

Note: if vcs not found error occured, try installing the development tools from the step 2. If it still doesn't work, try pip3 install vcstool.

6. Add the following lines to ~/.bashrc

alias rosbiped='source ~/biped_ros2/install/setup.bash'
alias rosdashing='source /opt/ros/dashing/setup.bash'
source /usr/share/gazebo-9/setup.sh

Note: remember to restart your terminal after changing ~/.bashrc. Alternatively you can run source ~/.bashrc to use the new bashrc config.

7. Build the project.

cd ~/biped_ros2
rosdashing
colcon build  --symlink-install
# The --symlink-install is such that instead of creating new files in the install folder, it just points to the files in the source folder, i.e. no copied files, easier to iterate changes

Conda

1. Follow the normal installation until step 6 (step 6 done), or step 1 done in simple installation
2. Create a new conda environment with python 3.6 (skip to 3 if already exists)

conda create -n <your_env_name> python=3.6

3. Install relevant dependencies

# Check if we are in a conda environment
which pip # should return something like /home/<user>/anaconda3/envs/<your_env_name>/bin/pip

# Check python version
python --version # If not 3.6.x, create new environment

pip install empy==3.3.2
pip install catkin-pkg
pip install lark-parser==0.7.2

4. Build

# Make sure conda env is activated when doing this
cd ~/biped_ros2
rosdashing
colcon build --symlink-install

Note: most likely there will be CMake warning messages like this for libssl and libcrypto

Cannot generate a safe runtime search path for target

For now it doesn't impact anything and there is no known solution to us on this issue, so it is ignored. If you can help fix this issue please submit a PR or raise an issue to discuss. Related discussions 1, 2

5. Run. Follow normal run instructions but with conda environment activated

Running

1. Open new terminal and run rosdashing and then rosbiped
2. Run ros2 launch arm_simulation debug_launch_all.launch.py to run the RRR robotic arm
3. Alternatively, run ros2 run openai_ros2 robot_arm_random_continuous to run the arm with random inputs

Updating

1. Switch to a version of choice (preferably latest)

cd ~/biped_ros2
git fetch --tags
git checkout <version_number> # version_number is something like 0.6

2. Run vcs import --recursive src < main.repos again to update the main repositories
3. Build the project again (colcon build)
4. Run

OpenAI Gym Integration

Refer to here

Notes

The biped stuff are not tested from version 0.3 onwards. Might not work properly.

Docker

Status

• Both spinningup and stable-baselines container is available through CD
  • Combined container still under consideration, but unlikely to happen
• The current implementation requires a GUI, and as such X11 forwarding is necessary to run the containers
  • Headless containers coming soon

Prerequisites

1. Running on an X11 system such as Ubuntu 18.04 using xorg
2. nvidia-docker

Running

Spinning up

Run the following command in a terminal. The python commands follow the normal spinning up running convention, documentation here.

xhost +local: && \
docker run --gpus all --privileged --rm \
  -u=1099 \
  -e DISPLAY=$DISPLAY \
  -e USER=defaultuser \
  -v "/tmp/.X11-unix:/tmp/.X11-unix" \
  -v "/home/$USER/:/home/$USER/" \
  --name=spinningup_container \
  pohzhiee/docker_sims:spinningup \
  python -m spinup.run td3 --env LobotArmContinuous-v2 --exp_name some_experiment

Stable-baselines

This requires a separate python script to run, an example code snippet shown below. Refer to the stable-baselines quickstart guide for more info.

Stable-baselines script snippet

env = gym.make('LobotArmContinuous-v2')
# Optional: PPO2 requires a vectorized environment to run
# the env is now wrapped automatically when passing it to the constructor
# env = DummyVecEnv([lambda: env])

model = PPO2(MlpPolicy, env, verbose=1)
model.learn(total_timesteps=10000)

To run this script inside the container, map the script into the container directory and run it.

For example, if your script is in /home/pohzhiee/Scripts/TD3_1.py, then run the following command

xhost +local: && \
docker run --gpus all --privileged --rm \
  -u=1099 \
  -e DISPLAY=$DISPLAY \
  -e USER=defaultuser \
  -v "/tmp/.X11-unix:/tmp/.X11-unix" \
  -v "/home/$USER/:/home/$USER/" \
  -v "/home/pohzhiee/Script/TD3_1.py:/home/defaultuser/TD3_1.py" \
  --name=baselines_container \
  pohzhiee/docker_sims:stable-baselines \
  python TD3_1.py

Notes

Robot arm environment strings

As of 17 Jan 2020 (version 0.6.10), the environment strings are as follows:

1. LobotArmContinuous-v0
   • Fixed goal, no random start, no state noise
2. LobotArmContinuous-v1
   • Fixed goal, no random start, with state noise
3. LobotArmContinuous-v2
   • Random goal, no random start, no state noise
4. LobotArmContinuous-v3
   • Random goal, no random start, with state noise
5. LobotArmContinuous-v4
   • Random goal, random start, no state noise

For further information, check the environment registration in the master branch here

Not enough privilege for docker run

• Run it with sudo, however this is not tested and as such the behaviour is not known

Transferring trained model to host computer.

1. If the training is not yet started, you can map the data output directory to your host through adding another -v flag. For example, if the data in the container is being saved to /home/defaultuser/spinningup/data/some_experiment/some_experiment_s0/, add -v /home/$USER/host_data_dir:/home/defaultuser/spinningup/data/some_experiment/some_experiment_s0/ as one of the run arguments

2. If the training has already started and no volumes other than X11 and home directories are mapped

   • Run docker exec -it spinningup_container /bin/bash
   • Once inside the docker container, run cp command to copy from /home/defaultuser/<some_path_to_data> to /home/<your_username>/<some_folder>
   • Example: sudo cp -r /home/defaultuser/spinningup/data/some_experiment/some_experiment_s0 /home/pohzhiee/advasdkvn/
   • Run exit to exit the container shell

3. Playing it safe (in case you forgot to map volume and the container shuts down before you can transfer anything)

   • Run without the --rm flag
   • After the container shuts down, check that the container still exists by running docker container ls -a
   • If the container still exists, run docker start -a -i <container_name> to start the container again (with previous data intact)
   • Note that this will run the training again, so try to copy the data before it gets overwritten
   • Run docker exec -it <container_name> /bin/bash to access the shell of the container
   • See how to copy in 2)
   • Because the container is still intact even after training ends, you will need to explicitly remove it through docker container rm <container_name before trying to create another container of the same name

4. Uploading to cloud server

   • Still under consideration, currently not supported