RoBits is a lightweight, modular and scalable software stack for AI-driven robotic manipulation. It is designed for seamless integration with robotic manipulation policies, by providing essential tools for perception and robotic control.
Why RoBits? RoBits features an intuitive command-line interface to get started quickly. It's user-friendly and ensures adaptability, efficiency, and ease of use across diverse robotic applications to help with research experiments.
RoBits comes with some default configurations, for example for the Franka Panda
robot. Run pip install 'robits[all]'. This will install necessary dependencies and
provide an entry point command rb. You can use rb to get a list of
commands. For example rb move home moves the robot to a default joint
position.
Checkout the Documentation for more details
Once you have installed RoBits you can access the CLI with rb in your shell:
(robits-py3.8) markus @ sockeye ➜ ~ rb [2025-02-23 19:53:43]
Usage: rb [OPTIONS] COMMAND [ARGS]...
RoBits command line interface for robotic manipulation.
This is the main entry point for all RoBits CLI commands. It provides global
verbosity controls and sets up logging configuration.
Use --verbose/-v to increase log detail level (can be used multiple times).
Use --quiet/-q to suppress all but error messages.
Example: rb -vv move home --robot-name robot_panda_sim
╭─ Options ────────────────────────────────────────────────────────────────────╮
│ --verbose -v INTEGER RANGE Increase verbosity of the logging output. Can │
│ base used multiple times │
│ --quiet -q Suppress all logging output except critical │
│ and error messages │
│ --help Show this message and exit. │
╰──────────────────────────────────────────────────────────────────────────────╯
╭─ Commands ───────────────────────────────────────────────────────────────────╮
│ camera Commands for working with cameras and visual data. │
│ config Commands for viewing and managing system configurations. │
│ dataset Dataset related commands │
│ gripper Gripper related commands │
│ info Read various information about the current robot state │
│ move Commands for moving the robot in different directions. │
│ panda Franka Panda related commands │
│ shell Creates an interactive shell. Use robot variable to interact │
│ speech Audio related commands │
╰──────────────────────────────────────────────────────────────────────────────╯To learn more about a command you can write for example rb dataset --help,
which will list the subcommands that are related to dataset.
You can display available config files with rb config list. To show the
actual content use rb config show <config_name> where config_name
specifies the name of the config file, e.g., rb config show robot_panda_real.
You can specify a user configuration by exporting ROBITS_CONFIG_DIR. E.g. by adding
export ROBITS_CONFIG_DIR=/home/markus/robits_configto your .bashrc.
This allows you to specify additional configurations for robots, cameras,
grippers. For example, you can change the serial number of the camera or the
ip address of a robot.
Once you set your user confiuration folder you can copy and modify an existing configuration with:
rb config copySelect the configuration you want to modify and open it with your favorite text editor
Best practice is to install a RT (Real-Time) kernel. See the script for more details.
With some patches, you can install the NVIDIA driver and access the GPU through
PyTorch on a real-time kernel system. This avoids unnecessary networking calls
and removes the burden of deploying the software as everything can be run from a
single repository.
Once that is done you can get the robot pose with:
rb info pose
To move the robot, use one of the rb move commands, e.g.:
rb move up
To test your camera setup you can use:
rb camera view and rb camera point-cloud view
For calibration to get the extrinsics, you can use:
rb camera calibrate extrinsics
This will launch a TUI. Select the camera and press "connect". Then use the
sliders to adjust the camera pose. Press "save" to store the camera calibration
to your user config directory, which is set by the environment variable
ROBITS_CONFIG_DIR.
Basic commands for the gripper are rb gripper open or rb gripper close.
Please note that there are some limitations with the Franka Panda gripper.
Basically, the robot stops when the gripper is active. See the implementation
for details.
To replay collected data, use rb dataset replay. You can specify the path, the control method, and the robot name:
rb dataset replay --input-path ~/data/demo_0000/ --robot-name robot_panda_sim --control-method positionBelow is a very simple example to initialize a robot in simulation and move it.
from robits.core.abc.control import control_types
from robits.core.factory import RobotFactory
robot = RobotFactory("robot_panda_sim").build()
with robot.control(control_types.cartesian) as ctrl:
ctrl.update(([0.52, -0.2, 0.15], [0, -1, 0, 0]))
ctrl.update(([0.0, 0.0, -0.13], [0, 0, 0, 1]), relative=True)
robot.gripper.close()See the Documentation for more details, such as command line integration, and for more examples.
If you are using this work or find it otherwise useful please cite:
M. Grotz, M. Shridhar, Y. Chao, T. Asfour and D. Fox
PerAct2: Benchmarking and Learning for Robotic Bimanual Manipulation Tasks.
https://doi.org/10.48550/arXiv.2407.00278
Also consider citing additional resources if necessary (See Acknowledgement)
- Libraries
- Real robots and grippers
- libfranka
- Franky
- pyRobotiqGripper originally from castetsb/pyRobotiqGripper
- xArm Python SDK
- Python URx
- Simulation