Hacker News Reader

Comment by perihelions | Hacker News Reader

How would you modify ant logic to fix this bug?

I would wager that, if you get a large ant mill going, the level of pheromone signal would become enormous and be sustained for as long as the mill is going and the ants are alive. So, a sustained, very high level of pheromones might trigger some kind of break-glass response where they invert their logic and seek _low_ levels of pheromone, seek if get _off_ of their track. Or perhaps they begin to erase the pheromone trail (if they can).

Which breaks the ant mill, but now you have to get back to having a functioning colony. Don't know how that would work.

respect for arriving at a local maxima and calling it a day

Idea: spend one neuron to track how often the ant thinks it's turning right (left). The input could be differential motion between sets of legs, or an internal accelerometer (if it exists). It takes very little storage to track an exponential moving average (just one persistent variable),

https://en.wikipedia.org/wiki/Moving_average#Exponential_mov...

If the average "winding number" goes very high over a period of some minutes, then you're probably moving in circles.

But then there's this failure mode: figure-8 loops...

I have seen figure-8 ant mills on YouTube, fwiw. It involved a "going underground" step, eg, go through a tunnel, pop out, turn and walk over the tunnel, turn and go back into the tunnel. I'll see if I can find it again.

Or ants on a Klein bottle.

What if an ant is trying to go up a steep mound in a spiral?

Then it's being an idiot. Ants can easily climb vertical walls (and trees).

They can even travel fully upside-down on my ceilings. Cliff overhangs are no obstacle for them.

following unrewarding pheromone trail for too long could trigger a timeout and walk in the random direction. Then search for a new trail. Even better if an ant could just return to the last visited anthill on timeout

I'm sure you will get enough problems with the definition of "unrewarding" and time control to make the change a net loss. How stable is the travel speed of ants anyway? Because trails do change form all the time.

https://en.m.wikipedia.org/wiki/Cycle_detection

The tortoise and hare algorithm seems well suited!

But tortoise and hare have much more complex brains!

How can you fix bugs and still have ants?

You don't squash the bugs... you just get the bugs to result in the desired behavior.

You are fixing the bugs, not eliminating them.

It's not a bug, it's an hymenoptera!

As a comic in all seriousness, though, I would imagine any search strategy that avoids a local maximum could be useful, such as other siblings have pointed out.

But I wonder if the ants even have a way of detecting they are trapped in a local maximum. If the signal detection is simply based on the strength of the signal, I don't know how they could detect the trap. If the pheromones of each ant were somewhat different, and if ants have memory -- maybe that's what would be required.

Otherwise if all ants just take random walks deviating from the signal, would you expect them to become isolated, or simply form a more elaborate ant mill whose position changes until the path to the nest is found or they become exhausted?

Can we determine if it's a bug instead of a worst case scenario for an optimal algorithm?

Make a random ant take a random deviation from the phermone trail every few hours

Wouldn’t be enough to form pheromone trail strong enough to cause other ants to follow.

At best it would just cause ant mills to slowly bleed ants into their surroundings where individuals are at a high risk of predation or simply becoming lost and starving anyway. At worst they follow a random path back into the ant mill.

If it's truly a "death spiral" aren't they all going to die anyway? The off chance that one breaks the loop and finds it back to the colony seems better than just infinitely going in a loop. But I'm no ant entomologist haha

You should watch the videos of these things. They have thousands of individual ants in them, sure having one or two break away and making it back is better than zero. But it’s better in the same way that zero rice falling out of bag is better than one or two grains falling out, zero is definitely better, but nobody bothers to pickup dropped grains of rice.

i would probably try to attack the root cause

Reply on news.ycombinator.com