Slashdot Mirror
Self-Repairing Spacecraft Uses Ant Logic
Elitist_Phoenix wrote to mention a New Scientist story about what could be the first steps towards a self-repairing spacecraft. From the article: "The team at CSIRO, Australia's national research organisation, is working with NASA on the project and has so far created a model skin made up of 192 separate cells. Behind each cell is an impact sensor and a processor equipped with algorithms that allow it to communicate only with its immediate neighbours. Just as ants secrete pheromones to help guide other ants to food, the CSIRO algorithms leave digital messages in cells around the system, indicating for instance the position of the boundary around a damaged region. The cell's processor can use this information to route data around the affected area."
that when this thing gets built that it won't do ANY of what it was originally intended to do but will wind up costing about twenty times more than originally budgeted?
Is it fascism yet?
So why do you paste the link to the porn into your post instead of just putting it in your sig?
"Honey, there's a Mars Probe carrying away our potato salad! I told you we shouldn't have picnicked near JPL."
Seriously, though:
"Other groups are developing impact sensor systems controlled by a centralised processor. But such systems would fail if the area containing the processor were damaged. So a distributed system could be much more reliable, says Bill Prosser of NASA's Nondestructive Evaluation Sciences Branch in Langley, Virginia."
That kind of seems like overkill. It's like "One processor is too risky, so we should instead have 100." Have 3 processors and 3 busses. If something can damage all 3, then the probe is F'd beyond all repair anyhow. You have to wire power to 100 processors anyhow if you do that such that a damaged power bus can still take out multiple panels. Weight is premium on probes, and 99 processors is not a very effective use of weight.
Table-ized A.I.
A system like this has a greater degree of redundancy built-in at the cell level than you could possibly hope to achieve with a centralised database; each cell is already checking the status of its neighbours, so for a square grid you have four way redundancy (the grid needn't even be regular, as long as individual cells know how many neighbours it should have). Also, a central controller needs a direct link to each cell, which would probably mean a complex and heavy wiring loom (introducing another point of failure); 5mm^2 of silicon weighs a lot less than (say) 5 meters of copper wire, multiply the difference by the number of cells and you're looking at a substantial weight saving for an active sensor system*.
There's also the interesting possibility that any spare computing cycles could be put to other uses: add the ability for cells to transfer data to other cells arbitrarily instead of by physical proximity and you're looking at the basis of a hardware neural net (I can't think of an immediate use for this, since I wouldn't advocate putting a mission-critical computer on the outside of a spacecraft, but then I am not a rocket scientist).
*Aerospace is the only field where "nothing" has a tangible dollar value: the more "nothing" you can put in a spacecraft, the cheaper it is to launch.
(Side note: I'm not logged in as I type this, and the script-filter word is "meteors". Coincidence? Perhaps...)
Blank until