A Python package for generating analytical inverse kinematic solvers for robotics manipulators. The generated solvers support many robots in Mech-Mind Robotics and have been extensively tested in real-world deployment. The name yaik stands for Yet Another Analytical Inverse Kinematics generator.
- Closed-form, analytical inverse kinematics for serial chain robots in python/C++
- Support robots with 7 DoF or more, such as Kuka IIwa Arm, Panda Robot, or arms of the Atlas humanoid
- A new factorized polynomial solution based on groebner reduce
- General 6-dof solution using linear matrix equations tangent of half angles
- Analysis to avoid degenerated (or conditional degenerated) solutions
The gallery contains generated IK solvers for tens of robots. Even if your robot is not included in the library, it might include a robot with the same structure and can be used after parameter modification. The file format is
gallery/
puma/
puma_ik.yaml
puma_ik_generated.cpp
puma_ik_generated.py
{robot_name}/
{robot_name}_ik.yaml
{robot_name}_ik_generated.cpp
{robot_name}_ik_generated.py
To use the generated C++ solver, you need to include the yaik_cpp_common.h file, which further depends on Eigen. Please refer to the yaik_cpp repo for an example.
To use the generated Python solver, you need to import python_run_import.py, which further depends on numpy.
To generate a new solver for a robot that is not in the gallery, you need to install python3, pyyaml, attrs and sympy on your system. Optionally, you might install sagemath, which contains a faster groebner reduce implementation.
Then, you need to express your robot in Modified DH format. A documentation regarding that can be found in here. Several examples can be found in this file. Alternatively, the modified dh model can be converted from chain model in this file, or URDF using a tool in another library.
After that, you can run ik_solve.py to generate the solution for your robot in YAML format. This step can take a LOT of time, as we perform offline analysis to find the simplest solution and avoid degeneration. Moreover, this step might fail to find a feasible solution (although the pipeline succeeded for all the robots we tested).
Finally, you can use cpp_codegen.py/py_codegen.py to generate solvers in C++ and Python, similar to the ones in the gallery.
This work was done during the author's employment at Mech-Mind Robotics. We have learned a lot from existing works, in particular IkFast and IKBT.
If you have a new challenging robot that cannot be solved, please email me.