Notebook Entry

AIO for September 2, 2014

Last week's accomplishments

Objectives from last week:

  • Finish implementing a tight matrix eval broadcasted loop to speed up the inverted pendulum direct collocation. -- Partially done: Speed benchmarks are done with C with Cython and Fortran 95 with f2py. The constraint eval computation speed was doubled and there is a projected 50X speed increase for the constraint Jacobian eval speed, but that still needs to be implemented. See this blog post for details: http://www.moorepants.info/blog/fast-matrix-eval.html.
  • Meet with Brad to discuss implementing direct collocation in Opensim on Wednesday. -- Done: I gave Brad an intro to Ton's direct collocation methods and we discussed what an Opensim direct collocation tool would look like.
  • Write outline for direct collocation system id paper. -- Done: https://github.com/csu-hmc/inverted-pendulum-sys-id-paper
  • Write two sections for the perturbed walking paper. -- Done: abstract, intro and start of "raw data" sections are drafted, https://github.com/csu-hmc/perturbed-data-paper/blob/master/paper.tex.
  • Figure out the basic topics and time allotments for the topics at the IEEE open source hack night. -- Done: Event is set for Friday September 12 and flyers have been made.
  • Email RUG, PUG, and NASA Python guys about the SWC workshop and cc Ahmet. -- Done: Tristan from NASA is asking around there about the interest level.
  • Write outline for a bicycle data paper. -- Done : https://github.com/moorepants/bicycle-data-paper

Other things I did:

  • Discussed visiting scholar project with Ton. We are thinking of creating a DC tool that will allow the user to supply system dynamics, additional constraints, and desired unknowns. It should be able to solve Ton's previous gait problems, my system id problems, and other general test cases. Ton likes the GPOPS API in Matlab which could copied for the Matlab wrapper. The tool would likely be implemented in C++ and wrapped for Python and Matlab use.

Last week's issues

  • Setup offline backups for the HMC website -- Still 90% done. I didn't work on this last week, overestimated time.
  • Run more example optimizations on the inverted pendulum (after speed up is implemented). -- Haven't finished the speed optimizations, plus goal is too broad. Breaking this one up more below.
  • Finish table in the steer torque paper. -- Overestimated time in the holiday week.
  • Add in torque wrench section into the steer torque paper. -- Same as above.
  • Go through the next two chapters in Accelerated C++. Overestimated time. Removing this one for now, till I get back to some C++.

This week's objectives

  • Fix the permissions issues for offline backups for the HMC website.
  • Implement the 50X faster constraints Jacobian evaluation for the pendulum id problem.
  • Finish table in the steer torque paper. Update the spreadsheet and insert into the latex doc.
  • Add in torque wrench section into the steer torque paper.
  • Write the draft of the raw data section in the perturbed data paper.
  • Add more detail to direct collocation id outline.
  • Write an introduction to the direct collocation id paper.
  • Fix all remaining issues on the yeadon paper and submit to F1000Research.
  • Create a figshare item for the raw bicycle data.
  • Touch base with Luke on visiting scholar idea.
  • Think of the exercises I will use for teaching git/github in the open source software night.
  • Advertise for open source software night.
comments powered by Disqus