This project implements most features from the original franka_ros
repository in ROS2 Humble, specifically for the Franka Emika Robot (Panda).
This project significantly expands upon the original franka_ros2
from the company, who dropped the support for the Pandas.
Additionally, multi-arm mujoco simulation has been integrated, meaning you can now run the same controller on both simulated and real robots.
Work is ongoing to integrate FR3 into the architecture.
As of 15.07.24, both the real and simulation robot interfaces are complete. GARMI (a robot platform with 2 panda arms and a non-holonomic base) is also available.
This repository is a copy of an internal repo, that we try to keep up to date. Unfortunately, it means it doesn't get nearly as much attention as it should. We plan to migrate fully over to this public repo in the near future.
Documentation for this project is available here.
More thorough information is available in the documentation.
- Real robot:
- FrankaState broadcaster
- All control interfaces (torque, position, velocity, Cartesian).
- Example controllers for all interfaces
- Controllers are swappable using rqt_controller_manager
- Runtime franka::ControlException error recovery via
~/service_server/error_recovery
- Upon recovery, the previously executed control loop will be executed again, so no reloading necessary.
- Runtime internal parameter setter services much like what is offered in the updated
franka_ros2
- Sim robot:
- Same as the real robot, except Cartesian command interface is not available, and there is no plan to implement this for now.
- Gripper server with identical interface to the real gripper (i.e. action servers)
- Example controllers for the real single-arm listed above, that correspond to those interfaces, work out of the box.
- FrankaState implements the basics: torque, joint position/velocity,
O_T_EE
andO_F_ext_hat
. - Model provides all the existing functions:
pose
,zeroJacobian
,bodyJacobian
,mass
,gravity
,coriolis
.- Gravity for now just returns the corresponding
qfrc_gravcomp
force from mujoco. - Coriolis =
qfrc_bias - qfrc_gravcomp
- Gravity for now just returns the corresponding
- Joint position controller might cause some bad motor behaviors. Suggest using torque or velocity for now.
- The default
franka_moveit_config
package depends onwarehouse_ros_mongo
, which has been deprecated. It has been changed towarehouse_ros_sqlite
for now to ensure thatrosdep install
works properly. For now, please refer to this discussion here for a potential solution; once a proper solution has been identified, it will be added.
- Build the docker image:
- at the root of the repository (where
Dockerfile
is located), runsudo docker build -t "bimanual:garmi" ./
, or whatever name you want; just adjust them in thedocker-compose.yml
or thedocker run
command below. - This installs all the required packages including
rosdep
.
- at the root of the repository (where
- Start the docker container with
docker compose
(for iGPUs):xhost +; sudo docker compose -f docker-compose.yml up
xhost +
is needed to give the Docker container access to the host's screen.- The
docker-compose.yml
file defines configurations that allow the container to run on the realtime kernel of the host, assuming that the host has one. The instructions were found in here. - Currently, running GUI apps outputs warnings in the console. Those can be ignored.
- Alternatively, with
docker run
(this has no RT kernel, so only the sim would work)xhost + sudo docker run \ -dit \ -v /tmp/.X11-unix:/tmp/.X11-unix \ --device=/dev/dri:/dev/dri \ --env="DISPLAY=$DISPLAY" \ --name=my_container \ --network="host" \ bimanual:garmi
- The
--network="host"
is what enables the FCI IP address to be found from inside Docker without any additional configuration. TODO: Add a more elegant way to handle this.
- The
- For other GPUs, please refer online.
- Access the container with:
- for
compose
,sudo docker exec -it realtime_humble bash
- for
run
,sudo docker exec -it my_container bash
- for
- (OPTIONAL) Check that the connection to the robot, RT kernel and screen are all working fine.
- You can check if docker is properly linked to your screen by running:
ros2 run rviz2 rviz2
or~/Libraries/mujoco/bin/simulate
- For RT kernel and robot connection, run
~/Libraries/libfranka/bin/communication_test <robot-ip>
- You can check if docker is properly linked to your screen by running:
(Tested on Ubuntu 22.04, ROS2 Humble, Panda 4.2.2 & 4.2.1, libfranka
0.9.2 and MuJoCo 3.1.6)
On a computer running Ubuntu 22.04 and real-time kernel (if you wish to use it with a real robot), do the following:
- Build
libfranka
0.9.2 from source by following the instructions. - Build MuJoCo from source by following the instructions (tested with 3.1.3).
- Install library dependencies:
- Install GLFW with
sudo apt-get install libglfw3;sudo apt-get install libglfw3-dev
- Install Eigen 3.3.9. Remove Eigen 3.4.0 if that was installed from following the
libfranka
steps.- Some functions will break if you use Eigen 3.4.0, and will fail to compile.
- Install
dq-robotics
C++ version
- Install GLFW with
- Install ROS2 Humble by following their instructions and create a workspace.
- Clone this repository (i.e. the multipanda) into your workspace's
src
folder. - Install the dependencies by running this rosdep command from the workspace root:
rosdep install --from-paths src -y --ignore-src
- Export the
{mujoco/libfranka}/lib/cmake
directories as aCMAKE_PREFIX_PATH
in your environment, i.e.- in
~/.bashrc
,export CMAKE_PREFIX_PATH={path to mujoco installation}/lib/cmake}:{path to libfranka installation}/lib/cmake
- in
- Source the workspace, then in your workspace root, call:
colcon build --cmake-args -DCMAKE_BUILD_TYPE=Release
- The paths to mujoco and libfranka are added to
CMAKE_PREFIX_PATH
, so no separate args are needed.
- The paths to mujoco and libfranka are added to
- Add the build path to your
LD_LIBRARY_PATH
by adding the following line to your~/.bashrc
:export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:{path to libfranka install}/lib:{path to mujoco's install}/lib
- The MuJoCo path is the INSTALLED folder's directory. The
lib
folder should only have the two.so
files, and a folder calledcmake
. - Likewise, the
libfranka
path should contain thecmake
folder and the.so
files.
- The MuJoCo path is the INSTALLED folder's directory. The
- To run:
- Single arm:
- with real robot, source the workspace, and run:
- Default:
ros2 launch franka_bringup franka.launch.py robot_ip:=<fci-ip>
. - Multi-mode:
ros2 launch franka_bringup multimode_franka.launch.py robot_ip:=<fci-ip>
. arm_id
is fixed topanda
.
- Default:
- with sim robot, source the workspace, and run:
- Default:
ros2 launch franka_bringup franka_sim.launch.py
. - Multi-mode:
ros2 launch franka_bringup multimode_sim.launch.py
. arm_id
is fixed topanda
.- Currently, only the robot with gripper attachment is available.
- Default:
- with real robot, source the workspace, and run:
- Dual arm:
- with real robot, source the workspace, and run:
- Default:
ros2 launch franka_bringup dual_franka.launch.py robot_ip_1:=<robot-1-fci-ip> robot_ip_2:=<robot-2-fci-ip> arm_id_1=<robot-1-name> arm_id_2=<robot-2-name>
. - Multi-mode:
ros2 launch franka_bringup dual_multimode_franka.launch.py robot_ip_1:=<robot-1-fci-ip> robot_ip_2:=<robot-2-fci-ip> arm_id_1=<robot-1-name> arm_id_2=<robot-2-name>
. - There is no default
arm_id
. - Grippers are set to true by default; add
hand_n=false
to disable this.
- Default:
- with sim robot, source the workspace, and run:
- Default:
ros2 launch franka_bringup dual_franka_sim.launch.py
. - Multi-mode:
ros2 launch franka_bringup dual_multimode_sim.launch.py
arm_id_1=mj_left
andarm_id_2=mj_right
by default.
- Default:
- with real robot, source the workspace, and run:
- Garmi:
- with sim robot, source the workspace, and run:
ros2 launch franka_bringup sim_garmi.launch.py
- with sim robot, source the workspace, and run:
- Single arm:
The original version is forked from mcbed's port of franka_ros2 for humble.
All packages of multipanda_ros2
are licensed under the Apache 2.0 license, following franka_ros2
.