View source

{{Languages}}
= ikfast =

Part of [http://openrave.programmingvision.com OpenRAVE].

Given any set of joints in any OpenRAVE robot, '''ikfast''' automatically generates closed-form 6D/3D inverse kinematics solutions for those joints given the position of the end effector of one of the robot links. It is not trivial to create hand-optimized inverse kinematics solutions for arms that can capture all degenerate cases, having closed-form IK speeds up many tasks including planning algorithms, so it really is a must for most robotics researchers. '''IKfast''' will hopefully be a useful tool for researchers throughout the world.

Closed-form solutions are necessary for motion planning due to 2 reasons:

* Numerical inverse kinematics solvers will always be much slower than closed form solutions. Planners require being able to process thousands of configurations per second. The closed-form code generated by ikfast can produce solutions on the order of '''~6 microseconds'''! As a comparison, most numerical solutions are on the order of 10 milliseconds (assuming good convergence).
* The null space of the solution set can be explored because all solutions are computed.

'''ikfast''' uses [http://code.google.com/p/sympy/ sympy] for the symbolic math and consists of only one python file: ikfast.py. A patched version '''sympy 0.6.3''' is included in any installation of OpenRAVE. For the curious, the patches applied are in '''ikfast/sympy_0.6.3_patches.diff'''.

== Features ==

* Generate inverse kinematics for 6 joints for an end effector rotation and translation (6D).
** At the moment, algorithm assumes that the translation and rotational components of the kinematics can be separated.
* Generate inverse kinematics for 3 joints for an end effector rotation only (3D).
* All possible discrete solutions calculated (usually up to 8)
* Generated C++ code '''independent''' of OpenRAVE or any other library.
* Automatically detects degenerate cases where 2 or more axes align and cause infinite solutions.
* Invalid solutions are detected by checking if square roots are given negative values or arc sines and arc cosines are given inputs exceeding the [-1,1] range.
* All divide by zero conditions are automatically checked and handled.

== OpenRAVE Usage (recommended) ==

OpenRAVE allows '''ikfast''' to be called for any robot using the [http://openrave.programmingvision.com/ordocs/en/openravepy-html/openravepy.databases.inversekinematics-module.html inversekinematics] database generator. It will generate the C++ code, compile it, and automatically link it with openrave. 

== Stand-alone Usage ==

It is possible to run ikfast.py by itself (it is in $OPENRAVE_INSTALL/share/openrave/openravepy) without OpenRAVE.

To get help and a description of the ikfast arguments type

 ./ikfast.py --help

Windows users will have to install python manually and add it to the environment path.

=== Forward Kinematics File ===

'''ikfast''' can be also used as a standalone program without openrave by providing a '''forward kinematics''' file for the robot. The '''KinBody::WriteForwardKinematics''' function outputs into this format. It is also possible to do this in a command line with openrave.py:

 openrave.py -p "open('fkfile.txt','w').write(env.GetBodies()[0].GetForwardKinematics()); sys.exit(0)" myrobot.robot.xml

=== Compiling and Using ===

In order to compile the generated c++ file in Linux/MacOSX do:

 gcc -lstdc++ -o ik_barrettwam ik_barrettwam.cpp

This will generate a small program that outputs all solutions given the end effector with respect to the robot base.

== Robots ==

Here are some robots with their auto-generated ik files:

=== Barrett WAM - 7DOF ===

[[media:ik_barrettwam.cpp]] [[image:barrettwam_robot.jpg|200px]]

=== Mitsubishi PA-10 - 7DOF ===

[[media:ik_pa10.cpp]] [[image:pa10_robot.jpg|200px]]

=== Puma Arm - 6DOF ===

[[media:ik_puma.cpp]] [[image:puma_robot.jpg|100px]]

=== Manus Arm - 6DOF ===

[[media:ik_manusleftarm.cpp]] [[image:manusarm_robot.jpg|200px]]

=== Man1 - 7DOF ===

[[media:ik_man1left.cpp]] [[media:ik_man1right.cpp]] [[image:man1_robot.jpg|100px]]

== Authors ==

[http://www.programmingvision.com Rosen Diankov] - PhD graduate student at the Robotics Institute at Carnegie Mellon University.