circa 1999, GTE Internetworking (which had been BBN and would be part of Verizon later) moved to a new office building in Burlington, MA. There were several design failures there (the carpet patterns had moire effects, the main corridor was curved so that you could not see what was up ahead) and the biggest failure was the automatic motion-sensing light switches in every office.

They were designed to save money, because nobody would forget to turn off the lights. They did an excellent job of keeping people in the dark, partially because of the above sensor problem -- typing on a keyboard doesn't provide enough movement to trigger the sensors -- and partially because of their incredibly high failure rate.

At about $50 per sensor module installed, they had to keep an electrician on full-time replacing sensors. I heard that eventually they stopped relying on trouble tickets and just swept the building over about a two week cycle.

Meanwhile, people bought lights that they could control by themselves. Many of them were halogens running on 150 to 450W.

Failed all the way around.

You could try rigging a cheap metronome to the sensor. Or even an led that lights every few minutes.

We managed to get hold of some kind of remote that's supposed to disable the sensor for two hours (though it's difficult to actually test, since there's no feedback at all when pressing it).

Since we're already using a Raspberry Pi for other purposes, I have the project of decoding the remote's protocol and driving an infrared led from one of the RPI's GPIO ports.

At Mozilla's mountain view office, the lights automatically turn off at night. To keep them on, you have to call a phone number and punch in what floor you are on. But of course this was annoying because you weren't always near a phone or aware of the number, so there is an intranet website that uses some VoIP service to phone the building controller to keep the lights on.

Motorised moving object on a microcontroller, perhaps?