System and Robot setup

This guide walks through installing and setting up RoBits.

Warning

It is highly recommended to use a real-time (RT) kernel as otherwise the time-sensitive control tasks may fail.

Single machine setup

Make sure that the system has booted the real-time kernel by running uname -a. To install a real-time kernel you can execute the script scripts/10_install_rt.sh
(Optional) You can install the Nvidia drivers while also using the RT preempt patched kernel. This is a little bit hacky, but seems to work. Either execute scripts/30_install_nvidia_dkms.sh or install it manually. See Nvidia DKMS for further details.

Multi machine setup

The following assumes that you have at least two different machines, one for control with a RT preempt patched kernel and the other machine used for inference with a GPU. Configure networking for inter-machine communication. Ensure:

Both machines are on the same subnet or can reach each other.
Use a wired connection for lower latency and greater stability.
Control PC: Install the real-time kernel on this machine. See scripts/10_install_rt.sh and Real-time Kernel
Perception PC: Install the Nvidia drivers. You can use the ones that are shipped with your distribution

Next install RoBits on both machines. Start the service on the Control PC with rb service start <name>. For example, rb service start robot_panda_real. Then copy and edit the robot_client configuration. Use this configuration on the Perception PC to connect to the Control PC. You can test it with a simple command such as rb move up --robot-name robot_client.

Real-time Kernel

See the Kernel wiki for actively maintained PREEMPT_RT versions that match your current kernel version. See the script scripts/20_install_rt_patch.sh for more details.

Nvidia DKMS

Note

Installing the Nvidia driver with DKMS is only required if you have a RT patched kernel.

If you are using a RT preempt patched kernel it is possible to install the Nvidia driver by specifying the IGNORE_PREEMPT_RT_PRESENCE=1 flag (See this gist).

Here is an alternative method using the Dynamic Kernel Module Support (DKMS). DKMS is a framework to automatically rebuild and install kernel modules when the kernel is updated. You can also use the script scripts/30_nvidia_dkms.sh

Install the linux-headers metapackage for your kernel flavour and the Nvidia DKMS package with

export NVIDIA_DRIVER_VERSION=560
sudo apt install linux-headers-$(uname -r)
sudo apt install nvidia-dkms-${NVIDIA_DRIVER_VERSION}${SERVER}

Then build the module and load it

cat <<EOF | sudo patch /usr/src/nvidia-560.35.03/conftest.sh
--- a/conftest.sh
+++ b/conftest.sh
@@ -716,8 +716,9 @@
 if [ -n "\$IGNORE_PREEMPT_RT_PRESENCE" ]; then
-    exit 0
+    echo Skipping preempt RT check
 fi
EOF

sudo sed -i '/if \[.*\$IGNORE_PREEMPT_RT_PRESENCE.*\]/,/fi/{
    s/^ *exit 0 *$/    echo Skipping preempt RT check/
}' /usr/src/nvidia-560.35.03/conftest.sh


dkms status
sudo dkms install -m nvidia/560.35.03 -j 12
sudo depmod -a
sudo modprobe nvidia

Verify that everything is working with uname -a and nvidia-smi. You should be able to see something like this

markus @ computer ➜  sam2act git:(master) ✗  uname -r                                                                                                                                     [2025-05-28 17:17:06]
6.8.0-59-generic

markus @ computer ➜  sam2act git:(master) ✗  nvidia-smi | head -n4                                                                                                                        [2025-05-28 17:19:43]
Wed May 28 17:19:50 2025
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 560.35.03              Driver Version: 560.35.03      CUDA Version: 12.6     |
|-----------------------------------------+------------------------+----------------------+

Franka Panda setup

Install the robot according to the manual. By default the webinterface of the Franka Panda is reachable at https://172.16.0.2. If you are using a different IP address make sure to update the configuration. If you want to use the CLI to unlock/lock the robot you need to specify the username and password in the configuration or create an rselab user with the robot’s web interface by clicking the “+” button.

Currently two different grippers are supported, the Franka default and Robotiq. Specify the one you are using in the robot’s configuration with the gripper keyword. If you are using a gripper other than the default make sure that you select the current end-effector other in the webinterface in the end-effector section and add the correct weight for compliance control.

Camera calibration

If you are using an Intel RealSense camera you need to extract the intrinsic parameters. You can use rb camera calibrate intrinsics for this step. To calibrate the extrinsic parameters run rb camera calibrate extrinsics. Select the camera in the drop down menu and press “Connect” button. To refine an existing calibration use the sliders to adjust the parameters and then press “Save” button. You can show the existing camera calibration either with rb config show or rb camera info extrinsics/intrinsics. You can display the current point cloud with rb camera point-cloud view as a sanity check. The coordinate system should be in the robot’s base with x-axis facing front.