Bio and Publications

Thursday, July 12, 2012

Innovation, Arduino and "Tinkering"

Many of my customers (mostly super-large IT shops) wouldn't recognize innovative behavior.  Large organizations tend to punish defectors (folks that don't conform), and innovation is non-conforming.

I've just had two object lessons in innovation.  The state of the art has left many in-house IT development processes in the dust.  The cost and complexity of innovation has fallen, but organizations continue to lumber along pretending that innovation is risky or complex.

You can find endless advice on how to foster a culture of innovation.  Often, this advice includes a suggestion that innovative projects should somehow "fail fast".  I'm deeply suspicious of "fail fast" advice.  I think it misleads IT management into thinking there's a super-cheap way to innovate.  It's misleading because "fail fast" leaves too many questions unanswered.

  • How soon do you know something's about to be a failure?  
  • What's the deadline that applies so that failure can happen quickly?  
  • What's the leading indicator of failure?

If you are gifted enough to predict the future -- and can predict failure -- why not apply that gift to predicting success?  Give up on the silliness of unstructured "innovation" and simply implement what will not fail.

At MADExpo, I saw an eye-opening presentation on the Arduino.  I followed that with viewing Massimo Banzi's TED Talk on the subject of Arduino, imagination and open source.

There are two central parts of the Arduino philosophy.

  • Tinkering.
  • Interaction Design.

Background

Without delving too deeply, I'm trying to build a device that will measure the position of a hydraulic piston.  It's the hydraulic steering on a boat, and a measurement of the piston position provides the rudder position, something that's handy for adjusting sail trim to reduce the strain on an autopilot.

Clearly, such a device needs to be calibrated with the extreme port and starboard range of motion.  Barring unusual circumstances, the amidships position is simply the center between the two limits.

Part 1.  Buy an Arduino, a Sharp GP2Y0A02YK0F IR distance measurer (for 10-80 cm), plus miscellaneous things like breadboard, jumpers, LED's, test leads, etc.  A trip to Radio Shack covers most of the bases.  The rest comes from SparkfunRobot Shop, Digi-Key and Mouser.

Part 2.  Learn the language (a subset of C.)  Learn core algorithms (de-bouncing buttons and the IR sensor).

Tinkering

At this point, we've tinkered.  Heavily.

What's important for IT managers is that tinkering doesn't have a project plan.  It doesn't have a simple schedule and clear milestones.  It's a learning process.  It's knowledge acquisition.

The current replacement for tinkering is training.  Rather than learn by attempting (and failing), IT managers hire experts to pass on knowledge.  This is, generally, limiting and specifically stifles innovation.

Years ago, I worked on embedded systems: hardware software hybrids.  We burned ROMs and programmed in assembler.  Back in those days, this kind of tinkering was difficult, and consequently frowned upon.  It was difficult to specify, locate, source, and assemble the components.  There was a lot of reading complex product data sheets to try and determine what to buy and how few were needed.

What had once been a very serious (and very difficult) electrical engineering exercise (IR sensor, button, LED, power supply, etc., etc.) was a few days of tinkering with commodity parts.  The price was low enough and availability ubiquitous enough that frying a few LED's is of no real consequence.  Even frying an Arduino or two isn't much of a concern.

Interaction Design

The next step is to work out the user interface.  For the normal operating mode, the input comes from the hydraulic piston and the output is some LED's to show the displacement left or right of center.  Pretty simple.

However.

There's the issue of calibration.  Clearly, the left and right limits (as well as center position) need to be calibrated into the device.

Just as clearly, this means that the device needs buttons and LED's to switch from normal mode to calibration mode.  And it needs some careful interaction design.  There are several operating modes (uncalibrated, calibrating, normal) with several submodes for calibrating (setting left, setting right, setting center.)

Once upon a time, we wrote long, wordy documents.  We drew complex UML state charts.  We drew all kinds of pictures to try and capture the important features of the interaction.

Enter Arduino

The point of Arduino is not to spend too much time up front over-specifying something that's probably a bad idea.   The point is to experiment quickly with different user interface and interaction experiences to see what works and what doesn't work.

The same is true of many modern development environments.  Web development, for example, can be done by using sophisticated frameworks, writing little backend code and messing with the jQuery, CSS and HTML5 aspects of the interaction.

The scales fell from my eyes when I started to document the various operating modes.

Arduino doesn't have a great unit testing environment.  It's for tinkering, after all.  It's also for building small, focused things.  Not large, rambling, hyper-complex things.  You can achieve complexity through the interaction of small, easy-to-test things.  But don't start out with complexity.

After writing a few paragraphs, I realized that the piston movements could easily be self-calibrating.  Simply track the maximum and minimum distances ever seen.  That's it.  Nothing more.  In the case of a boat, it means swinging the wheel from stop to stop to define the operating range.  That's it.

A button (to clear the accumulated history) is still useful.  But much simpler since it's a one-time-only reset button.  Nothing more.

Moving from idea to working prototype took less time than writing this blog post.

Next steps are to tinker with various display alternatives.  How many LED's?  What colors?  LCD Text Display?  There are numerous choices.

Rather than wasting times on UML, specifications, whiteboard and diagrams, it's a simpler matter to write the user stories and tinker with display hardware.

Tuesday, July 10, 2012

Cool stuff I saw at MADExpo

A random list of cool things.

  1. HTML5.  Start now.  It's supported (more or less) by all browsers if you add appropriate shims.  Start with sites like http://www.html5rocks.com/en/ and continue reading.  It's largely arrived and there's no compelling reason to delay.
  2. JavaScript.  I'm not a fan of the language.  However.  It's clearly here to stay. It's part of many important technologies (like CouchDB).  HTML5 shims (or shivs) are necessary.  Flex (and other browser plug-in languages) are effectively dead.  JavaScript is all that's left.
  3. Redis.  Wow! Is that cool?  Rather that fart around trying to get outstanding performance out of a clunky old RDBMS, use a simple, high performance data store.
  4. MongoDB.  Now that I've seen it, I have a vague notion of places where Mongo is better and places where Couch might be better.  In 80% of the application space, both are fine.  But there's a 20% where we might be able to split a hair and leverage the slightly different feature sets.
I don't build mobile apps.  But if I did, I think I'd start with Appcelerator Titanium before switching to a "native" development environment.  The idea that several standard features are readily accessible with a convenient development environment is pleasant.

I finally laid eyes on an Arduino.  Let me simply say that it looks like dangerous, dangerous fun.  I think I'll be hanging around at Radio Shack a lot.

Tuesday, July 3, 2012

Book Deal Fell Apart (sigh)

After spending a couple of years (really) working with a publisher, the deal has gone south.

The problem was—likely—all mine.  The book wasn't really what the publisher wanted.  Perhaps it was close and they thought they could edit it into shape.  And perhaps I wasn't responsive enough to criticism.

I like to think that I was responsive, since I made almost all of the suggested changes.

But the deal fell apart, so my subjective impression isn't too relevant.

What To Do?

I have two essential choices here.
  1. Finish up.  Simply press on with the heavily-edited version.  The path of least resistance.
  2. Roll back the content.  This involves going back to a less heavily edited version and merging in the relevant changes and corrections from the writing process.
Option 2 is on the table because my editor was unhappy with "digressive" material.  The editor had a vision of some kind of narrative arc that exposed Python in one smooth story line.  My attempted presentation was to expose language features and provide supporting details.

Perhaps I'm too deeply invested in the details of computer science.   Or.  Perhaps I'm just a lousy writer.  But I felt that the digressions were of some value because they could fill in some of the gaps I observed while coaching and teaching programmers over the last few decades.

In addition to the editorial challenge, there's technical challenge.  Do I step back from LaTeX?
  • Use LaTeX.  This means that I would have to create the non-PDF version with a LaTeX to HTML translator.  Read Converting LaTeX to HTML in the Modern Age.  See Pandoc.  Or, it means that I don't offer an HTML version (a disservice, I think.)  Also, I need to unwind the publisher's LaTeX style libraries and revert to a plain LaTeX.  Since LaTeX is semantically poor, I need to rework a lot of RST markup.
  • Revert to RST.  While RST tools can make both LaTeX and HTML from a common source, it does mean that some of the fancier LaTeX formatting has to go.  Specifically, I really got to know the algorithm and algorithmic packages.  I hate to give those up.  But maybe I can work something out with Sphinx 1.1.3's various features.
The problem is that three of the four combinations of paths have advantages.
  • Finish up using LaTeX is easiest.  Remove the publisher's document style.  Use HeVeA or Pandoc to make HTML.  Move on.  
  • Finish up but revert to RST.  This means conversion from the publisher's LaTeX to RST followed by some editing to fix the missing RST semantic markup and then some debugging to get the LaTeX and HTML to look good.
  • Rollback the content using LaTeX.  This would be challenging because I would have to merge manually edited publisher-specific LaTeX from the heavily-edited version with the Sphinx-generated LaTeX from the less-heavily edited source.  Things like the publishers' style tags would complicate the merge. 
  • Rollback the content and revert to RST. This means using Pandoc to convert the heavily-edited LaTeX to RST. Then merging the relevant edits into the RST original text.   This actually seems pretty clean, since the heavily-edited RST (converted from LaTeX) would be short.
Perhaps, if I was a better writer, I wouldn't have these problems.

It appears than the solution is MacFarlane's Pandoc.  This can reverse LaTeX to RST, allowing easy side-by-side merging of texts from various sources.  Or.  It can convert LaTeX to HTML, allowing easy work with the heavily edited LaTeX version.