Senses Necessary for Insect Flight

[Knowledge Seeker]

TL;DR Summary

It is possible to give a robot many "senses" like hearing, seeing, touching and so on. We could combine all those senses in a learning neural net system so that it can "learn" to navigate in space.

Which senses (hearing, seeing and so on) would be necessary to accomplish what flying insects do?

 

[anorlunda]

Which senses (hearing, seeing and so on) would be necessary to accomplish what flying insects do?

Are you writing a SciFi story?

Is your question about the biology of insects?

Why did you post it in the EE forum?

 

[FactChecker]

If you are also interested in basic flying, you should include the gyroscopes and accelerometers that allow an insect to determine its own motion. You may want to assume that that is an entirely separate subject, but I don't think you can completely ignore it. Imagine that you are close to a wall and want to stay away from it. If you feel that you are falling toward the wall, you would want to react, possibly even before you can see it.

 

[Borek]

you should include the gyroscopes and accelerometers that allow an insect to determine its own motion

Yep: https://en.wikipedia.org/wiki/Halteres

 

[sysprog]

It would be extremely difficult to emulate the flight capabilities of dragonflies -- ref: http://tolweb.org/accessory/flight?acc_id=2471

 

[Baluncore]

Remember the first law of the butterfly;
“The shortest distance between two points is a zigzag line”.

Insects survive to reproduce when they avoid predation by birds or bats. For that reason they have a built-in dither that makes their path unpredictable. Do you intend to include that dither in your drone?

Flying is precocious, insects do not need to learn to fly. To a large extent, the neurological system of an insect is pre-wired. For that reason, a plastic neural network that learns is not really an asset, it may be a liability.

You need to identify a raison d'être for flight before you can direct the necessary features that must be supported.

 

[Borek]

Flying is precocious, insects do not need to learn to fly. To a large extent, the neurological system of an insect is pre-wired. For that reason, a plastic neural network that learns is not really an asset, it may be a liability.

At the same time birds do need to learn (at least partially, apparently they do have some inborn reflexes). Actually even (rather primitive) autopilots used in RC multicopters and planes do "learn" - which mostly means fine tuning model parameters to better fit particular airframe.

I guess what I am trying to say is that there are many ways to skin that cat, I would not reject any ideas.

 

[essenmein]

Summary:: It is possible to give a robot many "senses" like hearing, seeing, touching and so on. We could combine all those senses in a learning neural net system so that it can "learn" to navigate in space.

Which senses (hearing, seeing and so on) would be necessary to accomplish what flying insects do?

Are you talking about using a ANN for the act of flying (narrow AI), or navigating in complex/unknown spaces (less narrow AI).

I think as far as flight is concerned, neural nets are there more to learn control parameters for a given drone, ie tuning the controller algorithm for a specific air frame rather than changing that algorithm to learn how to fly from first principles.

As far as navigating unknown spaces etc, there is a lot of work happening with things like drone swarms for mapping buildings etc. This is more challenging, esp if drones are small and cheap, ie not large number of fancy sensors.

Quick google search:
https://www.sciencedaily.com/releases/2019/10/191023172112.htm