bardhh/cbfkit — reverse-engineered prompt

Build me a Python robotics toolbox for safe robot planning and control using Control Barrier Functions. I want it to help a robot move toward a goal while avoiding obstacles, pedestrians, other robots, and uncertainty, with simple examples people can run right away.

Please include simulations for a unicycle robot, multi robot coordination in 2D and 3D, stochastic and disturbed motion, MPPI style planning, and a safety filter that can wrap continuous Gymnasium environments so an RL policy’s actions are made safe before they reach the simulator. I’d also like a neural CBF example where the barrier is learned from safe and unsafe samples.

Use JAX for fast math, automatic differentiation, and compilation. Include plotting or animation examples, benchmarks for the QP solver, tests, and clear tutorials so researchers can adapt it to their own robot models. Make the install process simple for Python 3.10 to 3.12, and look up current docs online if needed.