Journal2003May
From Simreal
Contents |
02 May 2003
My last journal entry, in the old system (which required me to edit the web page and upload it) was in November of 2002. That's six months ago, and yeah, I'm pathetic.
Not to mention still broke and busy. The CAD/CAM biz doesn't have the good sense to go under, but neither is it paying off like it could be. I blame, ummm, someone. Definitely.
Now that this journal is in the TWiki, I can edit it any time with just the click of a button. True, the input dialog and the button are hidden from you (bwa-ha-ha!), but that doesn't make them any less real.
If you, the crazed reader, are interested in a comment forum of some kind, e-mail me at edwin at simreal dot com (work out the proper format yourself, you are clever) and ask. Or heck, just MAKE one. This is a TWiki, after all.
Finally, I've decided to break with blog tradition (as best I can tell) and put these entries in chronological order... so you have to scroll to the bottom to see the most current entry.
02 May 2003
Okay, I think everything is switched over now. My poor TWiki server! It sits in my living room sucking DSL bandwidth. Ouch. I'll try to serve any larger files off of simreal.com.
In addition to my struggle to survive, I am about halfway through my fourth book. It is about artificial intelligence techniques, illustrated with Java code. It should be quite good! I may wimp out, however, and avoid a number of the more annoying statistical techniques lik Bayes and Kalman. Or maybe not. We'll see.
Strangely enough, I also harbour dreams of writing fiction. Again, we'll see.
09 May 2003
There are questions that I imagine my readers asking... they don't, of course, because people tend to not ask casual questions of authors. The actual questions I get are more technical, things like "what is up with that Fuzbol programmer?"
There are two questions in particular that I imagin. One is, why did you go and use Fuzbol in Applied Robotics and then not in Applied Robotics 2? The other is about the style of the books.
Fuzbol
Fuzbol came about because, on the one hand, I liked the AVR MCU used in that book, but on the other hand I couldn't find any cheap (read, "free") languages for it. And I didn't want to do the projects in assembly, not all of them. So, hey! Let's make a nifty fuzzy-based control language! It will be neat! I could promote it and sell it and stuff!
Right.
Turns out that Fuzbol is not a bad language, but the constraints on my time and energy kept it from getting upgraded until it was great. I don't regret writing Fuzbol, and I may work on it again someday... hopefully I'll be able to recoup the time and effort that went into it! Until then, it sits on my tax returns as Simreal's one big writeoff.
In the second book, I found a nifty 8051 MCU from Cygnal. Of course, the 8051 has more tools available for it than you can shake a stick at, so I didn't feel the need to port Fuzbol to it. And, the second book spends less time in the MCU than the first.
If I do a third robotics book, I'll probably use the JStamp MCU. Mmmm, native Java in an MCU. I'm not at all religous about my MCUs; they all do a job, and programming is pretty much programming.
Style
In my current book I talk about the "three books theory". For any given topic I feel you should have three different books that discuss it. One book never really gives you all the details you need, but between three different ways of describing something you can usually make it work.
My books so far are not written as reference works, or the ultimate last word on anything. In fact, I purposely chose a style and form that is different from the other books on the market to try and fill in gaps rather than compete directly. For example, Gordon <nop>McComb does a great job in his style of book and I see no need to reinvent that wheel.
Even though the books are technical, I take more of a story-telling process, where I start at one end and work my way through different projects and techniques until we pop out the end. They are meant to be read and enjoyed as inspiration as well as used for their projects.
The first half or so tend to be practical projects and, by the end of the book (and the end of the deadline) they have much harder projects that I don't describe completely. In AR2 the last few chapters are just discussions in AI.
I think of the Applied Robotics books as launching points. So when I do a bunch of stuff on an MCU but don't get down into the details, it's to show what can be done. There are lots of books on microcontrollers, so I don't feel compelled (because of my three-books theory) to give every little detail. You can find what you need easily enough. But then you can come back to my book and make my project. It's still there.
Because of this, my books may come across as schizophrenic. On the one hand there are easy projects laid out in detail. On the other hand, there are hard projects that are not fully explained. If you hate it, that's okay! If you like it, then you are clearly who I am writing for.
The current AI book is different, however. More about that another day.
11 May 2003
Todays nifty website: Strindberg and Helium
19 May 2003
Well, I just bought the tickets so my son can visit me during July. It sucks to live two thousand miles away from your son, though that's pretty much my own damn fault. Moved to Texas in 1999. He's in Oregon.
It will be good to see him, though. He's 14 this year, getting to be big (and I'm getting to be old).
31 May 2003
Last day of May! Woo! Next week I need to find an easy way to archive this month and move on to the next.
Working on the AI book the last two weeks ... ever since I submitted chapters 3 through 5 on the 15th... the bit on reinforcement learning has been kicking my ass. It all seems so simple! Well, duh, like the ambitious fool that I am I picked a particular involved example.
The plain vanilla pole-cart simulation, but instead of a simple partioning of the state space I went with a fancy hashed tile-space scheme to learn the continuous reward surface. Or something. Its based on the CMAC neural net thingy from the University of New Hampshire, http://www.ece.unh.edu/robots/cmac.htm .
Anyway, the example works, it learns, it makes pretty phase-space pictures... it just won't keep the pole up for more than about five hundred steps, with occasional spurts of 3,000 step balancings.
But I have to go on...
One book, ten chapters, about three weeks per chapter to organize the notes, write the example code, and write the text itself. Why do I keep doing this to myself?
31 May 2003
So, ever since September of 2002 or so I've been taking ballroom dancing lessons. This started as a wedding gift from a friend of my, the talented ballroom dance teacher Richard Fowler. The excuse was to prepare me for the inevitable first dance ... inevitable because Marla is a long-term dancer and there was no way I was escaping that ritual.
Turns out I enjoyed it.
Tonight we are going out dancing for the first time since the wedding. Two lessons a month without any floor practice makes me barely competent to show my face in public, but heck, it should be fun.
