Usage¶
Running the server and robot client (simulation)¶
Run both server & client on your machine:¶
To run the server and robot client for Franka:
launch_robot.py robot_client=bullet_sim use_real_time=false gui=[true|false]
launch_robot.py is an executable script which starts both a controller manager server and a robot client which connects to it, and is installed as a script with the conda package by default in $CONDA_PREFIX/bin. It utilizes Hydra configurations to easily swap between simulation and hardware.
For simulation, the gui option controls whether the PyBullet simulation renders, and requires an active display (i.e. you may need to use VNC if sshing).
Running the server and robot client on Franka Panda hardware¶
Run the server on the NUC:¶
Wait for your robot to boot up (wait for the blinking yellow light turns solid yellow).
If your robot has its joints locked (yellow light), unlock it from Franka Desk.
If your robot’s external activation device (EAD) is not released (white light), release it (blue light). (If you disconnect from the robot for any reason, you need to re-release the EAD.)
Then:
# On the NUC
launch_robot.py robot_client=franka_hardware
For hardware, use_real_time acquires the necessary sudo rights for real-time optimizations (see here for the specific optimizations.)
For debugging and calibration purposes, we provide a read-only mode for the robot client, where no torques will be applied to the robot but robot states are still recorded normally. To launch the robot client in read-only mode:
# On the NUC
launch_robot.py robot_client=franka_hardware robot_client.executable_cfg.readonly=true
Running user-code controllers on your machine:¶
Once you start an instance of launch_robot.py, you can send commands by utilizing the example user code, found in scripts.
For real-time uses (e.g. on hardware), it is highly recommended that this is run on a different machine than the machine directly controlling the robot, to avoid interruptions of the real-time loop. To do this, simply edit the ip while constructing the RobotInterface, e.g. in the scripts.
To debug custom TorchScript policies, TorchScript can directly script print statements, which will output on the server (i.e. for hardware, it will print on the NUC machine, where the launch_robot.py script was launched).
Running the gripper on your machine¶
We currently support two grippers: Franka Hand and Robotiq 2F Gripper.
To run the gripper server:
launch_gripper.py gripper=<franka_hand|robotiq_2f>
launch_gripper.py launches a service that exposes gripper functionality to a connected GripperInterface.
You can modify the configuration on the command line through Hydra. For example, to change the comport (communication port) while using the Robotiq 2F gripper config:
launch_gripper.py gripper=robotiq_2f gripper.comport=/dev/ttyUSB1
Using the Allegro Hand Client¶
The Allegro Hand client communicates using the SocketCAN interface which is included with modern Linux kernels.
Before starting the Allegro Hand client you must configure your CAN netdev interface and specify the CAN bus network interface name in the allegro_hardware.yaml configuration file.
One way to configure the CAN netdev interface is:
sudo ip link set can0 type can bitrate 1000000
sudo ip link set can0 up
you can use the command:
ip link ls
to check the names of the available CAN interfaces in your system.
To launch the Allegro Hand client run the command:
launch_robot.py robot_client=allegro_hardware robot_model=allegro_hand
Note: if you previously installed the PEAK pcan kernel modules for libpcanbasic, the SocketCAN drivers have likely been blocklisted. On Ubuntu systems you can undo this by commenting out all lines in the file /etc/modprobe.d/blacklist-peak.conf