Reverse-engineering insect brains to make robots | Hacker News Reader

Reverse-engineering insect brains to make robots

eetimes.com

121 pointsJoachimS34 comments

This reminds me of Rodney Brooks' (et al) work on Subsumption architecture, which I find quite enjoyable.

A subset of the AI group at MIT and also some non-MIT people (eg, R.Beer, L.Steels) during the 1980's produced many [similar] papers that I really enjoyed.

Really, ALL of the autonomous mobile robot people working with subsumption were pretty fantastic.

(Also, check out the Roomba vacuum lineage)

http://people.csail.mit.edu/brooks/papers/ -- "Elephants..." is a simple one.

Edit: (adding) If anyone in the Seattle area wants to chat about this, I can talk your ear off. I'm up for coffee/beers.

My experience has been that this architecture is really overrated. Sure, you can make an "instinctive" robot, but does it do anything useful, or just zoom around randomly like an insect? Robots are built to perform useful tasks that typically need to be precise and predictable. Two of the biggest categories of robot deployment are "lawnmower"-style operation (surveys, farming, etc.) and factory automation; what use is an unpredictable platform?

Roomba's gotten its ass kicked in the market by competing products that actually map and plan, instead of chaotically bouncing around and constantly missing spots on the rug. iRobot came around to doing (some) of the same thing, but it took them years to admit that it was the right approach.

Overrated? I believe that is appropriate if using subsumption by-itself as a higher-reasoning task.

I think it is more useful at foundation/primative-level controls with other architectures above it. Eg, I want the brakes in a vehicle orchestrated by some "world-overview"/3,000-meter view but the low-level stuff handled with direct-feedback of actuation and sensor pairings.

Today I learned:

> iRobot Corporation is an American technology company that designs and builds consumer robots. It was founded in 1990 by three members of MIT's Artificial Intelligence Lab, who designed robots for space exploration and military defense.

> The company's products include a range of autonomous home vacuum cleaners (Roomba), floor moppers (Braava), and other autonomous cleaning devices.

The founding members are: Rodney Brooks, Colin Angle, and Helen Greiner.

https://en.wikipedia.org/wiki/Rodney_Brooks

https://en.wikipedia.org/wiki/Colin_Angle

https://en.wikipedia.org/wiki/Helen_Greiner

A perfect venue for this kinda thing was http://dorkbotsea.org/ now it looks like it might be https://seattlerobotics.org/

Brooks and his team developed an embedded version of lisp for robotics called L.

As far as I know, it was never released publicly, though they published a paper describing it.

Do you have more info on L?

https://www.researchgate.net/publication/2949173_L_--_A_Comm...

http://www.chuckrosenberg.com/pubpg/index.html

L mentioned here: http://people.csail.mit.edu/brooks/papers/AIM-1227.ps.Z or http://people.csail.mit.edu/brooks/papers/AIM-1227.pdf

Also, see this discussion: https://news.ycombinator.com/item?id=28831353

Years ago I've read this about the state of the art knowledge on how human brain works:

"We have no idea how human brain actually works, there are billions of neurons. So instead we study brains of simpler organisms. Flatworm's brain has 52 neurons. We still have no idea how it works."

Does this still hold true or did we moved forward since?

The problem with the flatworm is that its brain can't do very much except move the body. There's no higher language and visual processing as in humans that you could compare the virtual brain against. So, if you want to actually verify the worm's brain in a simulation, you have to build the entire worm and simulate every aspect of it in a physically correct environment - ideally on a level of individual cells. This is a pretty hefty task, but the OpenWorm project is trying to do something like that: https://openworm.org/

As far as I know, we understand "wiring" of neurons, but we are probably missing "firmware" of neurons and that's the reason why real Flatworm is behaving differently from simulated Flatworm. It is not executing same program.

We can also ask if neurons can actually be simulated with current silicon based computer technology or if different type of computer is needed - Computer supporting operations from quantum physics? Chemical based computer where decisions and calculations are done by folding proteins?

There is still circuitry that we don't fully understand, along with the role of glia in information processing and with mysteries surrounding the processing capabilities of synapses/receptors and neuronal oscillations. That's not even the half of it.

There is so much in the way of biochemical pathways such as regulation of neurotransmitters and receptor and other protein trafficking and modification that the complexity is insanely difficult to comprehend.

And I haven't even mentioned the neuroendocrine system.

I realize I'm an uninformed layperson possibly spewing nonsense but is there any way we can scan the real flatworms' neural output as they dig, move, etc. and then labeling the data to train ML models?

I recently read a long essay by Stephen Wolfram in which he repeatedly suggested liquid-based computation as one possible way forward from current solid 3D substrates.

"No idea" is a bit strong, we actually have some amount of knowledge on the workings of the human brain, see https://www.researchgate.net/publication/51920832_Cognitive_...

We have general understanding of how some specialized cells work but no idea on how higher level abstractions exist (e.g. how visual input is transformed into spacial structure of the environment and how that is mapped into semantic understanding of a scene).

BEAM Robots [1] were some of the first robots I ever created in the late-90s. They were extremely accessible to someone still scraping together allowance and riding their bike to radioshack for parts. The photovore was one of the earliest designs I copied. It always astounded me how much the simple analog circuit "brain" made the robot's behavior appear complex and bug-like.

Stripped down to the bare essentials, the simplest photovore circuits [2][3] work by setting up a voltage divider between two photoresistors positioned like eyes on either side of the robot's body. The output is fed into an inverter based oscillation circuit biased by the side of the robot light is coming from. This runs through some more inverters in parallel to step up the current high enough for two tiny motors with wheels. It creates a robot that waddles towards the brightest light in the room.

The simplicity of the circuit makes the various emergent robot behaviors that much more surprising and interesting. The frequency of the oscillations get faster in brighter light and slower in dim conditions, almost like the robot is searching harder to find the light in a dark room. If one wheel gets stuck, the torque stalling motor backfeeds into the control circuit and it can sometimes jolt or vibrate itself unstuck. If you power the whole thing with a solarpanel that charges up a capacitor for bursts of runtime, the robot really seems to come alive as it wiggles it's way towards the light imperative that keep it's circuits fed. There is something magical when you see a robot struggling to stay alive.

[1] https://en.wikipedia.org/wiki/BEAM_robotics

[2] http://solarbotics.net/library/circuits/pix/4_comp_pvore.gif

[3] http://solarbotics.net/library/circuits/bot_popper_240.html

Lede :

" Opteran has successfully reverse–engineered the algorithm honeybees use for optical flow estimation (the apparent motion of objects in a scene caused by relative motion of the observer). This algorithm can do optical flow processing at 10 kHz for under a Watt, running on a small FPGA. "

If anyone from eetimes.com is reading... I'm looking for a column in EETimes from the late 70s. The gist of it was that there are three rules for achieving personal success in business: make money (i.e. at least break even, to keep it sustainable), have fun (being bored & miserable is not success), and work with people you trust (to peel away stress and open up opportunities).

If it's digitised and in their archives, chances are you could find it through various search operators on google(inurl will help a lot if they have dates in the url, etc).

I have tried at their site and at google. No luck.

But what's the chance that they have digitised stuff from the 70s ? Pretty low, I'd think. And FWIW, every issue in those days was BIG.

Insects have long been an inspiration for robots, and as someone who likes insects and robots, I support initiatives like this. However, one lesson that has been drawn from studying insects is that it's not all about the brain. A lot of the adaptive behavior of insects comes from the way their bodies are built. For example, cockroaches are very good at climbing over all sorts of rugged terrain that might pose problems for similarly designed legged robots, but a lot of that success comes from how well their legs are designed, not from a super sophisticated 3D model of the terrain or motion planning algorithm.

Seeing spiders were recently discovered as having REM sleep (dreams), this could be a surprise route to developing consciousness in machines.

Spiders can be pretty smart too it seems, at least for someone with only 600 000 neurons.

Portia have instinctive hunting tactics for their most common prey, but can improvise by trial and error against unfamiliar prey or in unfamiliar situations, and then remember the new approach.

They are capable of trying out a behavior to obtain feedback regarding success or failure, and they can plan ahead.

https://en.wikipedia.org/wiki/Portia_(spider)

I guess every predator profits from replaying past hunting experiences to achieve more success in the future, and basically that's what dreaming is, so it makes sense that spiders can dream. But I'm not sure if that's really a sign of consciousness?!

For there to be a dream there must be a dreamer, right?

Sure, as long as you're not trying to imply that the dreamer need be anything more than a machine that is now running in offline mode.

Happy to see another Black Mirror episode coming into fruition! :)

The headline also gave me some Westworld vibes :)

I'm surprised nobody mentioned Rick and Morty [1]

[1] https://www.youtube.com/watch?v=JwEubkHY8-I

Instead of the path of turning a machine into a sentient being, this approach is to turn a sentient being into a machine.

Applying "sentient" to insects is debatable... but yeah, I guess the first step to simulate a sentient being is to simulate any kind of being, even a relatively simple one (although if you look at the amazing things ants or bees can do, you realize that even insect brains are very capable things)...

Don’t piss off the forklift! It will try to sting you. /s

I wonder if we may accidentally carry over evolutionary behaviors.

I'd like to see mech battles where those mechs are controlled by insects. Just make a huge mech version of the insect. Spider vs. cockroach.

Giant combat robots controlled by insects. What could possibly go wrong?

Reply on news.ycombinator.com