A few new items have come in over the past week or so, and actually recording one of my "bigbot"’s night missions (i.e. when the roads and footpaths are clear of kibbizters and their pets) is now green-lighted.

Some of the recent devices include a couple of IR-assisted wifi cameras. For 20 usd they aren’t TOO bad, bad at least one suffers from the problem of "hum" over the audio circuit.

It seems the PWM used to power the IR lamps manages to make its way over the wifi to the destination at about 3db below (i.e. "not much, really") the actual noises of interest. But the 4Mpx image is nice and sharp, so I can’t complain there.

The wifi has a 200 mw transmitter so should be able to go the full 200 m of my block (don’t want to be crossing any unofficial dragstrips during the early am’s, either).

Both units have a bit of a drain on a 12v supply — 5-10w each, mostly for the lamps, of course.

So it may have been wiser to just have an IR-sensitive camera and an external (compatible) IR lamp mounted on a tripod, or somthing.

Anyway, I’m set to record mission 3 of the Ford, which is slated for sometime next week.

After a couple of experiments and a few mis-starts, I figure I’m ready to circle the block on full auto.

Mission 2 went off relatively smoothly around 1 wk ago.

After the mishap during mission 1 (go the 200 m to the "end" of the block) — where the truck took badly to a crack in the footpath pavement and proceeded to dismount/mount a couple of gutters to the detriment of a few connectors and wiring — I re-vamped the "edge of footpath" detectors and installed some modified (extended range) side-looking sonars.

Instead of a Kalman filter (things I’ve been working on lately) I returned to something I actually understand — Baysean filters — and made the sonars’ mission one of keeping the bot centered on the cement part of the footpath.

The method is pretty simple — keep the fenceline on the left of the bot an average of 12.5 m away, and the fenceline on the right an average of 1.5 m away. The fact that fences, trees, cracks in pavement, and gutters give various spurious and null returns is the reason for the filtering.

Much like a KF, the idea with the BF in this context is to decide — based on all the ranging returns we’ve seen to date (e.g. their average and the last 2 samples on each side) — where the Hell are we most likely to be in relation to the assumed footpath.

The idea uses all the a priori knowlege that’s around — i.e. the width of the road, footpath, and location of council-certified fencing wrt property lines.

Mission 2, therefore, was to move only about 50 m along the footpath, trying to stay in the centre using only the sonar ranging data.

Moving in ultra-low gear (you can;t say I don’t learn from my low-grade disasters), 15 m later the mission was successfully concluded with no problems.

This time, the broken parts of the concrete were negotiated OK, with the sonar data even providing stable heading info to allow for basic navigation around the problem area(s).

---

I have managed to get down to the toy sales. But, so far, I’ve only picked up a couple of minor things to amuse the cat. E.g. the Wowwee "tribot".

On sale it was 109 aud, but not really worth even that. On examination, it only really has the motors to drive the 3 omniwheels, plus a couple of servos to wiggle the eyebrows and ears. None of the limbs move — they are just "posable".

Anyway, the cat was amused with the "guard" mode.

---

Elsewhere, I’m still waiting for my "surplus trackbot" (seen prev) to turn up. It’s coming from Russia, so it’s possible there will be forseen unforeseen holdups.

Meanwhile, I’ve found yet-another-source for "bigbot" bodies. Short of going to full RC of a VW chassis, there is always stuff on ebay. E.g. this $300 highway paving machine: