Journal97 dec
From Simreal
Contents |
December 1997
Mon, 1 Dec 97
Well, I had a nice successful programming session tonight. Got a working Faketrode model working, that mimics the basic Adaptrode through an additive algorithm. Look for it on the BAWB area!
I also realized that I had put up my Robot Intelligence document in Word 6.0 format... and that this would fail for most of you readers. Almost, but not quite, worth an emergency update. Next update I will convert it to HTML or RTF... something generic.
Tue, 2 Dec 97
Argh! I went to the Boris page today to check the counter (I watch it compulsively... I get an average of 14 hits a day since I started caring) and found it lost a hundred hits. Hmmm... called customer support, they had a counter crash and backed up an old version. Bummer. Iím going to have them set it forward to where I expect it to be, get back in synch.
Now, to create a simulation of a single neuron with multiple Faketrode synapses...
Wed, 3 Dec 97
I want to create a nice, graphics-oriented program where you can build nets of Faketrode neurons and simulate them... but I donít have the luxury of working on this project for more than a couple-few hours a day. So instead, tonight I whip up a boring, text-oriented program to create nets of neurons and test them. Iíll keep the same graph output I have now and see where it leads me...
Fri, 5 Dec 97
Well, hmm, where was I? Oh yeah, I was going to build me a testbed for the Faketrodes, just for jollies, and to get a feel for them. Anyway, havenít done anything for a couple of days... just hid in my bedroom reading old Heinlein stories. I have never found it easy to work during the Holiday Season, and I was walking too close to the edge of industrial-grade burnout for my liking. I think Iím back now!
I found a nice article on Self-Organizing maps, too... this brain stuff should be fun.
Sun, 7 Dec 97
I dabbled a little bit on neural sims over the weekend, nothing heavy.
But look at the time! The year is about to clock over to a new digit, leaving me with about seven months left to build a working robot. Where am I?
I learned a stack about pneumatics, and built a couple of test legs that led to a design that may or may not work. I found that I donít have as much strength as I suspected I would, and a well-thought-out letter from Luc Bos helped me identify where extra losses could be coming from. If Boris canít do a decent walk (at least a ripple gait), then I need room for another hardware re-work near the end.
I have a thick (thick!) stack of AI and robot brain papers all around my desk, but have just a small dabbling on the simulations. Iíve been moving slow since Thanksgiving, and I honestly donít expect an energy pick-up until after Christmas or the New Year... the holidays have always been a laid-back season for me. The time I recharge and take a look at what is happening...
... and what I need is a schedule. But not yet. The rough plan is (an overview for my benefit at least, to collect my thoughts).... 1. Finish the basic technology research for the brains. A couple of weeks, maybe?
2. Test the brains in a small, rolling robot. Another month?
3. Build the new Boris hardware. This is fairly simple, but will require a stack of bills for the supporting valves and cylinders. February?
4. March, then, I want to be writing test code in a corporeal Boris. April at latest. That is the same month I will know if Boris needs a hardware re-design.
Four months? Hah! Double it. Puts me past August. Gotta shake booty! But then, as the line grows near, as the preliminary software starts to do something, excitement will build and replace frustration (unless, of course, it fails to work entirely)... and the pace will quicken.
There is still hope. I hope!
Now, back to the brains....
Oh, by the way, sorry about this weekís update. Itís fairly minor, but Iíll have a stack of good stuff next weekend. Promise! New links, new software, and new notes and references.
Thu, 11 Dec 97
Well, the net simulator works. Doesnít read or write files yet, but that is just a detail. It isnít very fancy, and it is limited to tiny little nets, but itís purpose in life is as a test of the technology...
... hmmm. But how do I test it? I need to supply data to it somehow, and watch how it reacts. I have a fancy towards Kohonen SOM nets, so I should make it possible to build a simple one of those. Needs sensory neurons... yeah, that can take their input values from a disk file (for easy repeatability). Maybe, in addition to the Constant, Learning, and Dumb neuron types, a Sensory type that takes as a parameter the file name. Iíll need to make the inputs slow enough to be able to watch the outputs. I should also add a randomize-weights function, so it doesnít start up blank.
Fun fun fun!
A slight problem in my Boris future... not technical, but practical. It looks like I may be going on retainer for a company soon, and that might do some serious damage to my time budget. However, they are also making noises like they want to sponsor Boris with some solid money, so my cash worries may end soon.
Sun, 14 Dec 97
Well, it looks like this is going to be the last Boris update of 1997. Between the Holidays, work, and another project that I have to focus on for a few weeks, Boris work is going to be mighty thin pickings.
So this update gives you a few more links, a few more journal notes, and a toy net simulator.
The Net Sim doesnít have much practical value, though I am using it get a feel for how different neural cell types work. I havenít gotten it to operate like a Kohonen net, Backprop, or anything else... that will have to wait until I can put more energy into it all. There are also a number of truck-sized UI holes you can drive through and crash it - so be careful! Neither does it read files as sensory neurons; almost better, though, is I added a full-featured expression evaluator in one of the neuron types. See the sample code. This is a toy for programmers. No documentation for users, sorry.
I debated for a while about putting up the source ... some of the code is from my generic toolbox of tricks, and I donít really want to give away the store. However, well, why not? Might help someone.
So you readers out there, and you know who you are, have a wonderful Holiday season, and a terrific new year. Iíll be doing a little Boris this and that, but the pages will be static until 1998. Maybe next week Iíll take the "New" markers off everything.
See you in the new year!
