So far no drones can do this. Even elaborate light shows rely on preprogrammed routes and deterministic behavior. This emergent, dynamic behavior where the individual birds all follow one another and a ‘flock’ comes to be is still beyond edge compute.

you can do this with like 20 lines of code. boid behavior is unbelievably simple

Accommodate all the motors and batteries being different + the momentums. Don’t just calculate positions, make sure each one gets to those positions. If you can make a drone swarm mumurate without input, you’ll become a billionaire/DOD will pay most of it.

I don't know what to tell you, other than I've programmed this sort of stuff 20+ times for all sorts of usecases. there really is nothing to it, it's just a sum of 4 different vector calculations (just 3 if you don't care whether they crash into the environment) given an array of neighbors w/in threshold radius and their distances. no ML, no AI, it's just like Game of Life if you're saying to rewrite the whole STACK for some reason including motor control, wireless nearfield comms, etc... ok. sure

Implement it for ten drones. Have it fly it up in the air with wind and show us how it performs.

you wanna put down the cash? I can give it a shot

10% finders fee for you if I get DoD'd

Crichton was on to something back in 2002.

Yeah : /

In fact that is pretty easy to do. This is an exercise you get in coding school. To make it there 3 rules to respect: 1If you get too near another bird, steer away from it; 2Try to fly towards the centre of the birds around you; 3Try to match the speed and direction of the birds around you. I think it doesn’t interest searchers to do it in reality as it has been simulated millions of times. While making drone fireworks requires way more brain power to achieve. But i would love to see the birds drone.

No. You are simulating point movements. You are not actually moving arbitrarily many things with different motive forces, battery capacitances, inertial moments for starting and stopping. Actually try modeling *that* out and report Big O complexity : )

Just try to run one of this well know simulation you will see by yourself, that exactly what birds do. Maybe you can add the battery dynamic, but that doesn’t change much

Birds do not have one big brain but act on their own and follow the 3 rules. It’s of course a simplification but pretty close to reality, that why it has been studied

This is not that difficult for computers to do. Saw examples of flocking over a 15 years ago after implementing “jellyfish”. The issue with drones is we’re still at high cost linear control systems like FPV, that are supply blocked. Not to mention there’s no demand for this, the current model gets on target well. But give it a generation. (<2 years)

No. It’s easy to compute where the positions should be ideally. It’s extraordinarily difficult to compute how to get to those positions.

Permit me some additional precision: Flocking is an emergent phenomenon not a computational one. Just because we can simulate an emergent phenomenon does not mean we can compute it. Computers can flock, ergo drones can flock. They don’t yet because they don’t need to.

@Indian_Bronson does compute = predict here?

@guimarin Yes, also update

@guimarin update as in, alter simulation in response to external inputs?

Yes, continuously. As an analogy/one aspect of the problem. All of those birds (and all the drones) are slightly different. Even manufacturing has tolerances. They have different motive forces, energy sustainment (battery capacitance), reaction times (network lag), etc. the complexity keeps increasing tremendously.

@guimarin The birds have brains and nervous systems purpose built (evolved) to do this, on learning epochs that are *extremely* long We still mis-classify spam lol

So far. Eventually they will be and things like individual and group navigation and actions without further input or communication will be something one to N many drones can do:

I wouldn't say it's beyond compute, we just don't know what they're computing. They might just have a small set of objectives that lead to this emergent behavior. Something like "I want to see another bird above me and the ground below me."

Yeah they have some set of operations that have been optimized into their brain and nervous system by evolution. Really hard (impossible) for us to mimic though (at present).

BOIDS algorithm has been around since 1986, but there would probably be a number of challenges in the real world.

It’s not enough

Solved in 1986 and even used in games like Half-Life.

But who will lead our ballistic bird swarms?

A cinematic dusk video of hundreds of drones flying like a starling murmuration, using Reynolds’ Boids rules (separation, alignment, cohesion) to form fluid, emergent patterns and adaptive waves in real time.

What’s the barrier for this? The neighbors theory is probably the most correct, relying on their speed &amp; velocity. But the drones are also relying on a wireless signal, which is delayed. Cracking that would be a major step.

Do it on paper and pencil/code notebook for a system with just two drones. Make sure you include inertia/momentum adjustments : )

What advantage would drones that can do that have over drones that can’t?

They could be given territories to simply patrol around in and coalesce and break up in groups dynamically as they need it wrt. other capacities (example: surveillance, pursuit, munitions deployment) without the pathing or individual behaviors ever needing to be specified or monitored by another system. You could just “let them loose” and they could do their own thing and move around in the world on their own volition.

That would look more like packs of hunting predators: wolves, lions maybe seagulls - unlike sparrow flocks, a lot more deadly and precise.

I feel like we're just months away from this, though.

It’ll happen, for sure

Why, its not like many swarming insects have particularly sophisticated decision criteria influencing their decisions. Is it mostly signal processing that is the barrier?

Signal processing but more importantly it’s not just a rule being computed to specify where the neighbor (or as a neighbor) one should be. It’s all the computations to *get* to that position.

This takes me back

murmurations be crazy

@0xdasha three dimensional networks

This is computable with decentralized rules. Each agent interacts with a topological neighborhood (about 6–7 nearest), not every agent. Three forces dominate: separation, alignment, cohesion. vᵢ’ = vᵢ + α(⟨v⟩{Nᵢ} − vᵢ) + β(⟨x⟩{Nᵢ} − xᵢ) + γ∑_{j∈Nᵢ} φ(‖xᵢ−xⱼ‖) + u_goal + ξ. Then vᵢ ← s·vᵢ’/‖vᵢ’‖. Small correlated heading changes propagate across the graph, creating global order. Near the order–disorder threshold, information travels far, which is why the group pivots as one. Tiny heading nudges spread by consensus, so a plurality bias sets direction. Not only is this not beyond the computer, it can be done right now.

Drones with a real time coordinated AI just might. Likely not too far out.

The most interesting aspect is that there is no "centralized cloud processing." The magnificent flock is created by free, fully independent individuals.

I used to read the old siggraph journals from the 80s in my college library back in the early 2000s. There was an old article about this that referred to “boids” (artificial birds). Still a great read:

Note the harmony results in formless form that carries a type of majesty. Drone formations are information, so can't penetrate beyond form into formless majesty of murmuration.

Everything is information

It’s part of what I’ve been (theoretically) working on, part of my Universal Momentum work. A lot more to do, but work in progress. (I’m not a drone guy, just a let’s-make-shit-better guy, and this falls into that)🫡

So far

That’s right - eventually, it’ll be possible. Once complex flock behaviors can emerge on top of preprogrammed rules (for example, locating power sources) there will be a whole new class of everything from surveillance drones to loitering munitions.

This is simply attractor dynamics, a classic N-point simulation in graphics. The reason they cannot do it is that drones are not collision-tolerant at all yet so it needs a grid of X distance to be safe. If they had 10k little drones with cages, it could be done.

There a super cool video by the esteemed Sebastian Lague called Coding Adventures: Boids. You need 3 rules: avoid crashing, steer in the same direction as those around you, and try to get to the center of whatever local group you’re in

The lost Tuna of the sky. The missing bird of prey.

We can never compete with nature. God is the ultimate designer.

playstation 3 homepage

Murmuration, one of my favorite words... drones would probably need some form of quantum sensing or entanglement with the inertial navigation units of the swarm to replicate it. Side note... do we need a new collective noun for herds of flying drones?

It’s mesmerizing.

Tbh I’m not sure this “compute”’as we understand it. I think there’s set of fairly basic rules within bird brains that allow them to reflexively engage swarm mechanics for flocks but group as whole might be brushing up against something in Morphogenetic Field we can’t measure yet. The existence of Psi phenomena hasn’t been properly taken into account in the development of AI systems and could easily be one of the primary reasons for so called “hallucinations”. The neurons in these bird brains are also subject to those same phenomena albeit at an extremely low level.

emergence

Yet

Birbs aren't real

why? we've had decentralized algorithms for this for decades. webGPU link below