Tutorial for Dynamic Walking 2006
Optimal Control Approaches To Dynamic Walking
Chris Atkeson, cga at cmu dot edu


I am currently looking for students who are interested in helping us get humanoid robots to behave with agility. General approaches are described here and some papers on walking are here.


Tutorial slides.


This page has pointers to relevant papers and software.


Example 1: Policy Search on Inverted Pendulum

Unpacking the zip file pol-srch-pend1.zip creates the directory pol-srch-pend1. cd Matlab to that directory. There are several .m files that you can execute: testsim, test1, test2, ... The notes.txt file has more information.


Example 2: Policy Search on Compass Gait Biped (Simplest Model)

Unpacking the zip file pol-srch-simple1.zip creates the directory pol-srch-simple1. cd Matlab to that directory. There are several .m files that you can execute: test1, test2, ... The notes.txt file has more information.


Example 3: Policy Search on Knee-ed Biped

This example is in C, but might get converted to Matlab by me or by you. It uses the Numerical Recipes in C library, which you must provide. Look at biped/biped/notes for more info.


Example 4: Trajectory Optimization on Inverted Pendulum

A very simple example of altering trajectory parameters.


Example 5: Trajectory Optimization on Inverted Pendulum II

Another inverted pendulum example of trajectory optimization, this time using splines. Executing the commands in notes.txt will lead to interesting results. This inverted pendulum is slightly different from that in Example 1 and 4.


Example 6: Dynamic Programming on Inverted Pendulum

This example is in C, but might get converted to Matlab by me or by you. Look at README.TXT


Example 7: Dynamic Programming on Compass Gait Biped

Thijs Mandersloot has kindly provided some example Matlab dynamic programming code for the compass gait walker (our version). The states are Poincare sections taken when the stance leg is vertical. Just run tm_sip_dp. The settings as they are now will run for three days. See if you can make this more efficient. Here is more info


Extra Bonus Stuff