Talos Whole Body Controller (WIP)

This is an attempt to create a simple Whole Body Controller from scratch. The controller is highly inspired from the works:

• Bouyarmane, K., Caron, Stéphane, Escande, A., & Kheddar, A., Multi-contact motion planning and control, In A. Goswami, & P. Vadakkepat (Eds.), Humanoid Robotics: A Reference (pp. 1763–1804) (2019). Dordrecht: Springer Netherlands.
• Salini, J., Barthélemy, Sébastien, Bidaud, P., & Padois, V., Whole-body motion synthesis with lqp-based controller – application to icub, In K. Mombaur, & K. Berns (Eds.), Modeling, Simulation and Optimization of Bipedal Walking (pp. 199–210) (2013). Berlin, Heidelberg: Springer Berlin Heidelberg.

This controller makes use of the excellent osqp QP solver and the osqp-eigen C++ interface.

Installation

Docker

This repository ships with a Dockerfile to create a fully functional image ready to be used. To create it, run this command while being inside this repo:

docker build -t ros_talos .

This will create an image with the name ros_talos. To launch it, do it with the script provided:

./run_container.sh

This script includes this repo inside the catkin workspace. That way you can modify the controller from the host. You may need to run these commands with sudo, depending on how docker is set up in your machine.

To access the container, open your browser and go to localhost:6080. You should be able to view the container’s desktop after a few seconds.

Read the Compiling and How to use sections to learn how to compile the workspace and run the controller inside a simulation.

Raw

The raw installation was tested on Debian Stretch (9), using ROS Melodic. Install the Talos simulation environment from the oficial ROS page. Also, you will need to install the Pinocchio library. To do so, follow these instructions. Also, do not forget to clone this repo inside your workspace!

Alternatively, you can have a look to the Dockerfile and use it as a reference, but that should not be necessary.

How to use

Compiling

Make sure you have the package python-catkin-tools installed. Then, in your catkin workspace, run:

catkin build --force-cmake -DCATKIN_ENABLE_TESTING=false

Simulation

The first step is to launch the Gazebo physics simulator:

roslaunch talos_gazebo talos_gazebo.launch

Controller

We can now load our controller:

roslaunch talos_wbc_controller talos_trajectory_wbc_controller.launch

The robot will remain still, waiting for a trajectory to follow. You can change the QP formulation parameters at any time using rqt:

rosrun rqt_reconfigure rqt_reconfigure

The controller is waiting for trajectories to be sent to the /talos_trajectory_wbc_controller/command topic. The message type is talos_wbc_controller::JointContactTrajectory, and can send joint and center of mass trajectories, as well as link contacts with the world.

Generate trajectories

Trajectories can be generated in advance using any planning framework. Those trajectories then have to be formulated as a JointContactTrajectory.

Using this fork of towr and the package xpp_talos, simple walking gaits can be generated and sent to this controller. This trajectory generator is already installed in the Docker image. If using a raw installation, you’ll have to clone it in the workspace and build it. To launch the trajectory generator environment, run:

roslaunch towr_ros towr_ros.launch

By default, the trajectory generated will have only two steps, one for each foot. So it is a good idea to have the goal moved near the initial position of the robot.

Every trajectory generated is automatically saved to ~/.ros/towr_trajectory.bag. To send the trajectory to the controller, use the node talos_contact_traj as follows:

rosrun xpp_talos talos_contact_traj

You should be able to see the robot trying to follow the trajectory in Gazebo.