PlaCo

Project notes

Choose the Python or C++ route

The reviewed README requires Python 3.10 or later and describes wheels for Linux and macOS on x86-64 and arm64. Its Python installation command is pip install placo; using the C++ API requires building from source through the separately linked instructions. The selected packet does not reproduce that build or establish availability for another operating system. An unpinned package installation is not necessarily the reviewed source revision. [1]

The README's first example loads placo.RobotWrapper("models/6axis/"), creates placo.KinematicsSolver(robot) and masks the floating base with solver.mask_fbase(True). The model path must exist and the assumption that the robot is anchored must fit the chosen mechanism. The external examples repository is where the README directs readers for robot models and scenarios; that repository and its assets were not inspected here. [1]

It then adds a frame task named effector, configures a soft task, calls robot.update_kinematics() and solver.solve(True). Before adapting this sketch, inspect the selected model's frame names and target pose: a string from an example does not establish that the frame exists in your robot. None of these commands was executed in this review. [1]

Separate solver state from physical commands

The sampled C++ kinematics implementation contains concrete task-building methods. For example, string-based frame tasks resolve frame indices, while a frame task combines position and orientation tasks. This supports investigating how the public interface maps targets into model tasks. It is a bounded implementation excerpt, not a whole-library audit or a reproduced solver result. [3]

The README describes a lightweight rigid-body implementation using Pinocchio as a reference, coal for collision calculations and qpmad as a QP solver. Its upstream validation statement is an author report; no comparison or performance test was rerun. Solving and integrating model state does not establish motor transport, feedback timing, joint-limit enforcement on a physical machine or safe recovery after a failed solve. [1]

PlaCo is MIT licensed; the README separately identifies the bundled qpmad solver as Apache-2.0. External robot models, meshes and dependencies retain their own terms. Begin with a matched model and one task, then inspect solver outcomes and configuration before extending the setup. No wheel installation, C++ build, model loading, collision check, QP solve, simulation or hardware-control test was performed. [1] [2] [3]