Operational Space Control Barrier Functions

Code for "Safe, Task-Consistent Manipulation with Operational Space Control Barrier Functions" -- Daniel Morton and Marco Pavone

Submitted to IROS 2025, Hangzhou

What is OSCBF?

This is a safe, high-performance, and easy-to-use controller / safety filter for robotic manipulators.

With OSCBF, you can...

• Operate at kilohertz speed even with over 400 active safety constraints
• Design safety constraints (barrier functions) easily via CBFpy
• Enforce safety on both torque-controlled and velocity-controlled robots
• Either apply a safety filter on top of your existing controller, or use our provided controller

In general, this will be especialy useful for enforcing safety during teleoperation or while executing learned policies

For more details and videos, check out the project webpage as well as the CBFpy documentation.

How do I use this?

Check out the examples/ folder for interactive demos in Pybullet! This is the best place to start

If you're applying this to a different robot, you'll need to provide a URDF -- we can parse the kinematics and dynamics from that. Some notes:

• I haven't written an MJCF parser yet, but it should be feasible
• All joints that shouldn't be controlled as part of the kinematic chain should be set to fixed (gripper joints, for instance). Since you probably want to still be able to use the gripper joints in simulation, the best way to handle this is to make a copy of the URDF: load the non-fixed one in sim, and parse the fixed one with OSCBF
• I manually defined the collision model for the Franka. There are probably better ways to parse or generate this data from meshes, but I haven't done it yet. For now, I'd recommend doing the same for your robot.

FAQ

• "I already have a well-tuned operational space controller! I don't want to replace that"
  • You can use OSCBF as a safety filter on top of your existing controller!
• "I'm working with a mobile manipulator. Does this still work?"
  • Yes! We support prismatic and revolute joints, so adding a mobile base just adds 3DOF (PPR) to the beginning of the kinematic chain.
• "I'm controlling joint-space motions -- does this still apply?"
  • Depending on the task, you will likely want to modify the objective function. But, the robot dynamics and CBF formulation will still be useful!

Installation

A virtual environment is optional, but highly recommended. For pyenv installation instructions, see here.

git clone https://github.com/stanfordasl/oscbf
cd oscbf
pip install -e .

Note: This code will work with most versions of jax, but there seems to have been a CPU slowdown introduced in version 0.4.32. To avoid this, I use version 0.4.30, which is the version indicated in this repo's pyproject.toml as well. However, feel free to use any version you like.

This has been tested on Python 3.10 and 3.11, on Ubuntu 22.04.

Documentation

See the CBFpy documentation, available at this link

Citation

@article{morton2025oscbf,
      author = {Morton, Daniel and Pavone, Marco},
      title = {Safe, Task-Consistent Manipulation with Operational Space Control Barrier Functions},
      year = {2025},
      journal = {arXiv preprint arXiv:2503.06736},
      note = {Submitted to IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Hangzhou, 2025},
      }