Notebook Entry


AIO for February 9, 2015

Last week(s)'s accomplishments

I didn't do an AIO for a while, so this is stuff since the last one.

Rando

  • Completed everything to be a SWC instructor and am now officially an instructor. I'm going to teach a workshop at UC Irvine in late February.
  • Had a PyDy sprint last weekend with 7 people to get the website and documentation cleaned up and digestible for newcomers.
  • Interviewed for an awesome robotics/machine learning job.

Yeadon Paper

  • Final review came in. The paper is accepted for indexing. We are going to submit a revision to address both reviewers' comments.

Perturbed Data Paper

  • [1 hr, Done 1.7] Create a web page on hmc.csuohio.edu for the paper.
  • Reviews came in. The paper is accepted with minor revisions.

Gait Direct Identification Paper

  • [1 hr, Not done] Setup a repository
  • [2 hr, Not done] Draft an outline

Quiet Standing Identification Paper

  • [5 hr, Done (too many hours)] Work on the ODE eval speed up for the shooting implementation.
  • [10 hr, Done] Write the code for running a single shooting method.
  • Sped up the integration significantly. For example a 10 second integration with 5000 requested time points runs in about 1.5s. I had to write an efficient interpolation method in Cython and I also tweaked PyDy's code generation stuff to gain some speed ups. This makes the data generation and shooting less time consuming.
  • All three methods now run: direct identification, indirect with direct collocation and shooting. For shooting, I can use any algorithm provided in scipy.optimize and additionally I have a parallelized CMAES implementation. Indirect with dc and CMAES can find a solution from an initial guess of zero for the gains. Other shooting algs are picky about the initial guess. DC is fast (5 s) and CMAES takes in the hour(s) ranges. Haven't gotten all the timings yet. SQLSP doesn't take long (several minutes) if you give it a pretty much perfect guess.
  • Some work on the bias from direct and indirect has been done (https://github.com/csu-hmc/inverted-pendulum-sys-id-paper/issues/1). I just need to run the indirect methods so I can compare.
  • I have a data generation routine setup that makes generating lots of data relatively easy.

Last week's issues

  • Still can't focus and get 40+ hrs of work done on research. I miss having to actually go to work.

This week's objectives

Yeadon Paper

  • [30 min] Email authors to start the second revision process and open some new Github issues.

Perturbed Data Paper

  • [1 hr] Create issues in the repository for all the fixes.
  • [30 min] Email co-authors with tasks and deadline.
  • [2 hr] Write draft of response in the github repo.

Quiet Standing Identification Paper

  • [6 hr] Finish direct/indirection bias comparison plot.
  • [5 hr] Finish computation speed results.
  • [5 hr] Make a gain accuracy vs duration of simulation plot.
  • [5 hr ] Try a set of initial conditions with each method and compare which ones converge and which don't.

Gait Control Direct ID Paper

  • [1 hr, Not done] Setup a repository (convert the walking-sys-id repo and move to csu-hmc org.
  • [2 hr, Not done] Draft an outline and what figures tables should be present.