Spine single-process control library

Project notes

Keep the historical Python range visible

The inspected metadata declares Python >=3.8,<3.11 and GPL-3.0-or-later, with numpy/pandas/scipy, TOML and plotting dependencies. Its installation route is a source checkout and editable install. The README explicitly says simulator and motor-driver libraries are not installed automatically; moteus, the selected T-Motor driver or PyBullet must be supplied separately. No backend environment was installed here. [1] [3]

A source workflow creates motors from a TOML robot description, creates a controller and passes both to the selected motor_control_loop. The controller derives from AbstractController. The mjbots loop is asynchronous, invoked as asyncio.run(motor_control_loop(...)); its TOML specifies model and motor_id. T-Motor adds bus_id alongside model/motor_id and requires the documented mini-cheetah-motor-driver-socketcan dependency. PyBullet instead needs a URDF path and joint indices. Sample values are examples, not validated settings. Preserve those backend-specific configurations rather than treating the shared call shape as verified device compatibility. [1]

An example output array is not a tested motor action

The documented controller returns a 3-by-m array of position, velocity and force/torque commands. Its minimal example returns recorded position and velocity with zero requested force/torque. That interface illustration is not verified inert behavior for every powered mechanism. The inspected mjbots loop contains actual command/recording logic to study alongside the selected actuator configuration. No motor, simulator, data recording or nominal control frequency was exercised. [1] [4]

The GPL project notice does not open bought-in actuators or replace dependencies' terms. Timing, torque limits, interruption handling and record correctness need tests in the selected environment; no real-time, persistence or safe-recovery claim was verified here. [2]