Notebook Entry

Today's task list:

  • [x] Send code for theano and sympy
  • [x] Work on parsing the walking data
  • [~] Figure out the bug I'm getting at the end of page 153 on the plone book
  • [] Work on the website theme
  • [] Figure out what to do about the incorrect budget category for the computers
  • [] Review the TODO items on the Yeadon paper
  • [] Do D-Flow/Cortex tutorial (Due August 12)

SymPy Code Generation

Another person needing code gen of complex SymPy derived ODEs: https://groups.google.com/forum/#!topic/sympy/VtaxCRNO4sE

Pushed my pydy code gen experiments: https://github.com/PythonDynamics/pydy-code-gen

And a patch for the theano stuff in sympy: https://github.com/sympy/sympy/pull/2358

Fred speed up my theano implementation a lot: https://github.com/PythonDynamics/pydy-code-gen/pull/1

Walking System Identification

I added a walk module to my DynamicistToolKit:

https://github.com/moorepants/DynamicistToolKit/blob/master/dtk/walk.py

For steady walking we need to derive the average motion (sorta the limit cycle). Steps:

  1. Use ground reaction force data to find heel strike and toe-off times.
  2. Splice the joint angle and joint rate data up according to the heelstrike/toe-off times. This will give stance phase of each leg, alternating.
  3. For the whole run average the angle, rate, and force data to get the average motion and torques.
  4. Now list the variables that are potentially feedback variables, $s$ and all of the potential control variables (joint torques), $M$.

Obinna gave me some new data with good ground reaction force measurements. The one Ton provided seemed like the person was dragging their left foot. Chris's heel strike/toe-off finder worked great for the data. I'm working up the maths for the controller solver and will post once I've scribbled enough in my paper notebook.

comments powered by Disqus