He has some great videos. I watched this one and was impressed at the calculations, the transparency and the celebration at the end when he reached supersonic speeds.
Or, in other words: There was some text to read and some pictures to look at. For topics in the range of "supersonic trebuchets", I'm absolutely okay with quickly reading some lines of text; dynamically deciding how quickly/thoroughly I read, what paragraphs to skip, etc. What I'll definitely NOT do: Watch a video clip about it.
So you won't support the creator, but support a person who summarizes the creator. You could have just pasted the link into Claude and asked for a summary of the blog post.
Maybe that's even the same guy?! But I don't know...
A thing that I completely don't care about: Supporting some YouTube guys.
YouTube "creators", even if they are not basically "influencers" (a.k.a. advertizing), even in the best case, a YouTube guy is for me the same as someone at RTL, DMAX, Fox, or any other random TV station. It might even be that I watch their show every now and then... But I'm not constantly thinking about how I can support these guys. And IF they actually disappear tomorrow, there will be other ones coming up.
There is DEFINITELY no scarcity in YouTube guys...
I suspect that with a little iterative modelling of the whole system with derivatives, efficiency could be bought from ~40% to ~80%, giving you an extra few hundred mph...
there might be several gaps before that, even. From the top of my head:
- text description to 3D geometry gap: you can't easily describe complex geometry in a language that's common and convenient to both humans and machines
- 3D geometry to physical model gap: material , geometrical constraints induced by manufacturing (machines, tools, costs, …)
- physical model to model fit for simulation gap: meshing, constraints, stress modelling, …
- simulation outcome to fitness assessment gap: now you have a high-dimensional and numerically heavy simulation to weigh against a non-rigorously defined acceptance criteria
- simulation outcome to geometry profiling gap: how do you even start to guide the LLM into the vast space of possible changes to apply to the 3D geometry and reboot that loop?
LLMs don't strike me as a particularly relevant technique to apply here, to be honest.
You can probably keep this kind of unconstructive comments to yourself. Even if this was a place to discuss one's beliefs, nobody was attacking yours to begin with.
I mean I can link a conversation with Claude that says it knows about materials and various details on their various strength but I don't think that'll convince you of anything.
Indeed, it won't: I can link a conversation with XYZ-bot that tells me that I'm handsome and my ideas are brilliant and I'm en route to tremendous success. The point is that LLMs are, by design and current models norms, sycophantic and incapable of describing their own knowledge and workings/processes. This is not a matter of "my opinion vs yours", that's what few minutes spent legitimately trying to understand how this works will tell you.
A LLM may produce OpenSCAD compliant syntax, I don't doubt that, but how do you make it produce designs that fit engineering constraints? How do you even express those to a LLM? How do you validate those? How do you iterate upon them?
The domain of ML-driven design optimization isn't exactly new, is quite specific, and I would need convincing that Claude has anything to contribute to it.
I’m pretty sure Claude has been trained on high school physics textbooks, yo. And if it hasn’t there are plenty of other models out there that have.
It’s not a problem of the ML model. It’s a problem of the human setting up the description of the problem in such a way that the model can operate. OpenSCAD gets a long way towards that target. Use it with a model thats been trained for the purpose - just the same way that ML has been used to produce optimal rocket engine nozzles and fuel transfer systems.
> Claude has been trained on high school physics textbooks, yo
The very fact that physics textbooks have little bearing in the real world is the whole damn reason why Mechanical Engineering exists as a separate discipline, yo
> It’s a problem of the human setting up the description of the problem in such a way that the model can operate.
It's a problem of defining the initial state (which is the trivial part that OpenSCAD may be a contributing element of), defining constraints and variables (what is allowed to be changed and not, for what can, in which ways, to what extent, i.e. what is the library of allowed material, fastening, machining, assembling techniques available to your very specific situation), defining evaluation and fitting criteria.
I see very little adequacy of general purpose LLMs in that
> the same way that ML has been used to produce optimal rocket engine nozzles and fuel transfer systems.
Which has nothing to do with "just use Claude, yo"
I'm pretty sure the only reason the video trebuchet wasn't a proper bullet is because it's made out of plastic instead of a ball bearing. Aren't most handguns subsonic?
This is nonsense. You're going to double the efficiency by using "derivatives". As if the expert in this video didn't do any modeling themselves (shown in the video) and wouldn't have considered acceleration at all.
I say, put your money where your mouth is and prove you can double the efficiency.
All the energy is lost to air resistance and left in the machine in the form of parts still moving when the projectile detaches.
Both of those can be modelled - rather precisely if you have the compute power. By modelling the derivative of all of those things with respect to every parameter of a parametric design, losses can be minimised with gradient descent and a huge Jacobian matrix.
A few parameters are obvious (string length, arm length, etc), but there is no need to keep the set small - you could model the exact shape of the projectile latch as a spline with 100 parameters for example, and let gradient descent find the best shape to minimize losses.
AI can set up and run all those simulations, and all you have to do is pay the AWS bill for all the open foam instances you'll be running.
I think the idea is that at the end the sling starts to unwind too, increasing the effective arm-length (that non-linearity is one of the reasons trebuchets are so efficient to begin with). As a result it's possible that decreasing the spool to near zero is like putting the gears of your car too high. Reducing the spool size ensures there's enough torque left to put more power into the final swing.
I suggest having a BBQ with your neighbors. That should minimize the pre-launch problems. If you use a steel ball bearing like he originally thought, that bullet will be the start of your legal problems.
Other trebuchet fans who dream of building one some day may be interested in googling "walking trebuchet" for a surprisingly high complexity/results ratio.
Every time I see a strandbeest[1] on the beach somewhere, I wonder to myself why it hasn't got an array of trebuchet included, also recharged by the wind, to pick up sand balls and throw at people.
I would love for this to be optimized to the point that its portable and can be carried around, assembled on the side of the mountain, and used as a low-cost, efficient way to throw things from one mountain to another, maybe for avalanche control, or cloud seeding, or indeed for seed bombs intended for re-wilding hard to access places, or something.
It just seems like such an obvious tool, rather than a weapon.
This is so obviously wrong. Show me a field-rechargeable gunpowder-based device for launching things, and I will show you someone who is about to run out of ammo on the side of a mountain they’ve just climbed after a single day of work.
Honestly, what the hell. Where do you think gunpowder comes from, thin air? Oh, is gravity a subscription service in your country?
Come on, try to at least be sensible about your response.
Guns are already ~20-50% efficient at turning chemical energy into kinetic energy. If you're not using a direct chemical reaction for energy, then you're using a human which is a lot less energy-dense than a backpack full of ammo.
And yes, people who do avalanche control typically fire explosives out of a cannon or drop them from a helicopter. Getting a machine supply all the energy is the whole idea instead of climbing mountains yourself.
Do you not understand your argument is exploring a made-up scenario where we're no longer an industrialized civilization where these things are manufactured cheaply by the billions?
And if you INSIST on making everything completely low-tech and renewable, build a potato gun and use high-proof alcohol as fuel. It'll weigh less and last longer than whatever rube-Goldberg trebuchet design you have.
>lift the weight for every shot, which requires energy.
All harvesting requires some secondary-energy source.
Gravity is a resource, it takes effort to convert it to a form that can be applied - just like wheat, photons, air guns, etc.
> It's not obvious whether a trebuchet is better than compressed air, like a BB gun.
True, it’s not obvious yet, but it sure is fun following along and imagining the final, highly optimal version making its way through the process. I’d use it for some seed bombing hijinks, anyway. Green up the place, I would.
Or if you have a limited supply of gunpowder but a way to make sufficient numbers of appropriately shaped projectiles, an air rifle would do the job and is much lighter and simpler than a trebuchet.
I looked into it. Yes, you're correct, I was wrong. Although there is a speed limit of mach 5 (theoretical, not reachable in practice!) for shockwaves in normal atmosphere.
Seed bombing is pretty fun. Re-wilding is a huge industry. Field-reloadable ultra-trebuchets that can be carried and deployed across the landscape, instead of stored in a weapons depot, also a big win for energy harvesters, everywhere.
Carrying an extra 50kg weights up a mountain seems inefficient to me, unless you mean a much lighter one or one that you can use stones you find around where you are?
Thinking this approach out, could someone correct me if I have this wrong.
So he gradually increases the energy transfer from the weight to the arm over the course of the drop. But that arm swinging around must have a lot of energy loss due to friction and air resistance.
This loss will rise exponentially with the speed increase right? So in a sense he is holding his energy budget in a very inefficient phase, with high losses.
Two things support this. A) he spent a lot of time trying to make the arm more arodynamic. B) The spool widening at the end, which is essentially changing back down gears, applying more torque, right at the end when you want speed.
An alternative would be to drop the weight, and let it fall unrestricted until it gets to it's highest speed (this would minimise friction and air resistance) and only at the end of its fall, transfer the energy into the arm.
In this case the arm would only need to rotate 180 degrees or whatever, and not waste energy rotating right?
And then, to take this a step further to make this work there would need to be some flexibility or elasticity in the system to take energy from the high speed falling weight to the arm.
And what you have ended up with is something that looks like.... A whip!
If I'm picturing what you're explaining correctly, I think this would be difficult. Suddenly accelerating the arm from zero to hundreds of mph would put immense stress on the arm, not to mention whatever you're planning on using to transfer that energy (you kinda hand waved that bit).
I think your assessment is correct about the energy losses. I'm just not sure about the fix.
I think the fix is essentially a whip, that's what I meant by elasticity and flexibility.
If you want to accelerate a weight quickly in a single swipe, (simplified) gradually reducing the mass along the length of the whip imparts more kenetic energy at the end (the whip crack).
So the arm would no longer be rigid, although you can see in the OPs video it isn't actually fully rigid.
The arm (or whip) would still have to accelerate from 0 to ~2000+ rpm on half a revolution even though its flexible. So there would still be enormous accelerations even though the whip eventually partly smoothens out the peak acceleration. it would be more like a collision which in turns introduces higher stresses (and probably vibrations too) to the system. Both requiring you to increase the strength + rigidity of the system and hence also increase weight on the arm. Which in turns increases inertia and the "collision" effect.
Not saying its impossible to create a successful system like that, in fact it sounds sort of a natural/organic design that potential could be closer to the optimal design, but it would be more difficult to accomplish and probably need more testing and design iterations. So I think OPs strategy is more simple and pragmatic in this case.
You'd need to build a large heavy whip, like a chain whip but those only work laying on the ground due to how heavy they are, because the mass at the very end of a whip gets going so fast /because/ it's extremely light compared to the rest of the whip. So scaling up the weight to even a few grams of projectile means the rest of the whip gets very heavy.
I strongly suggest you get into trebuchet building! It is a fascinating mixture of art + science. You can start with legos or a few popsicle sticks, paint sticks + screws, etc.
To make things easier, start with an "integrated sling ballistic" (ie: a piece of thread between two lego pieces). Sling release is actually fairly complicated so chop it out!
The arm is generally a rigid lever, and the sling acts exactly as your "gradually reducing mass". In his design you can see the kindof bike gear shape at the axle and eventually coming like an "A" towards the tip (and reducing to "infinite thin-ness" via the string/sling).
After you've built one simple (small) trebuchet, test to failure! How much weight can you stack on it before a component breaks (bucket? arm? axle? sling? frame?). How heavy of an object can you throw before "everything starts messing up"?
My recommendation for ballistics is mini-marshmallow (desktop), regular marshmallow, mega-marshmallow, then switch to "hard" ballistics, eg: mini candy pumpkin (the connoisseurs choice), tennis ball, basket ball, etc.
Once you start throwing dense, solid objects (marbles, golf balls, baseballs) your risk of property damage and injury goes up fairly exponentially, so stick to marshmallows and tennis balls until you feel more comfortable.
Spend lots of time figuring out safe/reliable trigger mechanisms, WITH A SAFETY! For my tennis-ball treb, I did three eyebolts with a dumb screwdriver on a string, and a vise-grip (clamping) on the far side. Eventually, you're trying to safely "release" 100's of pounds that is being held in tension, and if you have a misfire (fail to release trigger) you're potentially approaching a "loaded" machine which is demonstrably "not working right". Lifting/loading 100's of pounds and then having an unlocked trigger (which _also_ may fail) is not safe for anyone.
If you remove the arm, is it still a trebuchet? Even the current design is pushing definition of trebuchet. That said, a gravity powered supersonic whip-a-pult would be pretty awesome :)
Energy loss from air resistance typically is quadratic, not exponential, with regard to velocity. So it’s worth much less than you’re arguing. And the arm has to rotate from what I can understand, so the end rotational velocity - and hence, rotational energy - would be the same?
Isn't drag either linear (proportional to velocity) or quadratic (proportional to velocity squared) depending on the scenario? But yes quadratic for things at this scale?
Drag is linear with velocity under Stokes' Law which assumes laminar flow and a very very low Reynolds number, and does not apply to ballistic projectiles flying through air.
Aerodynamic drag on the kinds of things that are involved in trebuchets is always quadratic with velocity.
>And what you have ended up with is something that looks like.... A whip!
Had a similar feeling come to me as I watched this video a few weeks ago. It made me really want to see someone make a whip-based trebuchet that uses a big spring like https://www.youtube.com/watch?v=PkQZU-CBwhk
Whips work because they gradually get lighter and lighter along their length meaning the same energy as it propagates down the length results in higher and higher speeds. Strapping a projectile to the end throws all of that off unless you make the whip extremely heavy to compensate.
Not sure if you watched the video, but the chain they're using is extremely heavy. I would think the projectile would be on the order of one of the single chain links at the end of the small side.
A 3d printed ball like in the OPs video would only be less than 3grams I imagine.
Either way, the weight of the whip is hardly important right, as you are just doing an energy transfer from the falling weight into the small ball, the lighter the whip the better, as long as it doesn't break. If the whip is heavier, it will soak up the energy budget in momentum
He could have also used some kind of chemical reaction contained within a rigid tube to transfer the energy to a projectile, but I think he just really likes trebuchets in particular.
He talks about reducing the drag by improving the aerodynamics of various components and got a 17.8% increase in it's RPM with just a leading edge profile and wrapping it in mylar.
The big issue in your proposed version is building an arm that is light enough to not absorb a lot of energy (it still needs to be rotating at the same speed as the progressive version in the video) but rigid enough to survive that sudden acceleration. My bet is you'd lose way more to the arm flexing and the added weight than you would the benefit of not dealing with aerodynamic losses.
So what you're saying is he should extend the drop height a bit to run a vacuum pump before engaging? Sounds like he's on the path to launch some satellites!
Tom is an amazing engineer. His journey in incrementally engineering a plane powered by compressed air has been awesome, both to see the iteration, but also to see his skills grow.
98 comments
[ 1.1 ms ] story [ 30.9 ms ] threadHe has some great videos. I watched this one and was impressed at the calculations, the transparency and the celebration at the end when he reached supersonic speeds.
A thing that I completely don't care about: Supporting some YouTube guys.
YouTube "creators", even if they are not basically "influencers" (a.k.a. advertizing), even in the best case, a YouTube guy is for me the same as someone at RTL, DMAX, Fox, or any other random TV station. It might even be that I watch their show every now and then... But I'm not constantly thinking about how I can support these guys. And IF they actually disappear tomorrow, there will be other ones coming up.
There is DEFINITELY no scarcity in YouTube guys...
- text description to 3D geometry gap: you can't easily describe complex geometry in a language that's common and convenient to both humans and machines
- 3D geometry to physical model gap: material , geometrical constraints induced by manufacturing (machines, tools, costs, …)
- physical model to model fit for simulation gap: meshing, constraints, stress modelling, …
- simulation outcome to fitness assessment gap: now you have a high-dimensional and numerically heavy simulation to weigh against a non-rigorously defined acceptance criteria
- simulation outcome to geometry profiling gap: how do you even start to guide the LLM into the vast space of possible changes to apply to the 3D geometry and reboot that loop?
LLMs don't strike me as a particularly relevant technique to apply here, to be honest.
The domain of ML-driven design optimization isn't exactly new, is quite specific, and I would need convincing that Claude has anything to contribute to it.
It’s not a problem of the ML model. It’s a problem of the human setting up the description of the problem in such a way that the model can operate. OpenSCAD gets a long way towards that target. Use it with a model thats been trained for the purpose - just the same way that ML has been used to produce optimal rocket engine nozzles and fuel transfer systems.
Not that difficult, really.
> Claude has been trained on high school physics textbooks, yo
The very fact that physics textbooks have little bearing in the real world is the whole damn reason why Mechanical Engineering exists as a separate discipline, yo
> It’s a problem of the human setting up the description of the problem in such a way that the model can operate.
It's a problem of defining the initial state (which is the trivial part that OpenSCAD may be a contributing element of), defining constraints and variables (what is allowed to be changed and not, for what can, in which ways, to what extent, i.e. what is the library of allowed material, fastening, machining, assembling techniques available to your very specific situation), defining evaluation and fitting criteria.
I see very little adequacy of general purpose LLMs in that
> the same way that ML has been used to produce optimal rocket engine nozzles and fuel transfer systems.
Which has nothing to do with "just use Claude, yo"
I say, put your money where your mouth is and prove you can double the efficiency.
Both of those can be modelled - rather precisely if you have the compute power. By modelling the derivative of all of those things with respect to every parameter of a parametric design, losses can be minimised with gradient descent and a huge Jacobian matrix.
A few parameters are obvious (string length, arm length, etc), but there is no need to keep the set small - you could model the exact shape of the projectile latch as a spline with 100 parameters for example, and let gradient descent find the best shape to minimize losses.
AI can set up and run all those simulations, and all you have to do is pay the AWS bill for all the open foam instances you'll be running.
I would think you want it decreasing to near zero to extract all the kinetic energy from the mass, leaving the mass stationary as it hits the ground.
- structural: There are height and size limits on things you can build without a permit
- trespassing: flinging objects onto others people property is illegal
- endangerment: any activity that can endanger other people is illegal
All of these three depend on how rural you live I guess. So in a suburban neighborhood I would assume they get involved very quickly.
Other trebuchet fans who dream of building one some day may be interested in googling "walking trebuchet" for a surprisingly high complexity/results ratio.
Just for the fun of it. Pure whimsy.
[1] - https://www.strandbeest.com/
It just seems like such an obvious tool, rather than a weapon.
Is gravity a subscription service in your world?
One launch location or multiple?
What frequency of launch?
How heavy are the payloads?
How good accuracy?
What transport methods are available?
How long should it operate without maintenance?
Is solar an option if we go electric?
Etc.
Problem is the muzzle velocity is pretty paltry right now. I think we are stuck behind the energy density of the power source
Also pretty sure several parts of the mechanism would need to be considered consumable if operating over a long period.
Honestly, what the hell. Where do you think gunpowder comes from, thin air? Oh, is gravity a subscription service in your country?
Come on, try to at least be sensible about your response.
And yes, people who do avalanche control typically fire explosives out of a cannon or drop them from a helicopter. Getting a machine supply all the energy is the whole idea instead of climbing mountains yourself.
The point is to avoid that entirely and harvest gravity.
Re-wilding teams already climb far and wide, and this would extend their reach without increasing their dependencies.
Ammo is a dependency.
And if you INSIST on making everything completely low-tech and renewable, build a potato gun and use high-proof alcohol as fuel. It'll weigh less and last longer than whatever rube-Goldberg trebuchet design you have.
It's not obvious whether a trebuchet is better than compressed air, like a BB gun.
All harvesting requires some secondary-energy source.
Gravity is a resource, it takes effort to convert it to a form that can be applied - just like wheat, photons, air guns, etc.
> It's not obvious whether a trebuchet is better than compressed air, like a BB gun.
True, it’s not obvious yet, but it sure is fun following along and imagining the final, highly optimal version making its way through the process. I’d use it for some seed bombing hijinks, anyway. Green up the place, I would.
So he gradually increases the energy transfer from the weight to the arm over the course of the drop. But that arm swinging around must have a lot of energy loss due to friction and air resistance.
This loss will rise exponentially with the speed increase right? So in a sense he is holding his energy budget in a very inefficient phase, with high losses.
Two things support this. A) he spent a lot of time trying to make the arm more arodynamic. B) The spool widening at the end, which is essentially changing back down gears, applying more torque, right at the end when you want speed.
An alternative would be to drop the weight, and let it fall unrestricted until it gets to it's highest speed (this would minimise friction and air resistance) and only at the end of its fall, transfer the energy into the arm.
In this case the arm would only need to rotate 180 degrees or whatever, and not waste energy rotating right?
And then, to take this a step further to make this work there would need to be some flexibility or elasticity in the system to take energy from the high speed falling weight to the arm.
And what you have ended up with is something that looks like.... A whip!
I think your assessment is correct about the energy losses. I'm just not sure about the fix.
If you want to accelerate a weight quickly in a single swipe, (simplified) gradually reducing the mass along the length of the whip imparts more kenetic energy at the end (the whip crack).
So the arm would no longer be rigid, although you can see in the OPs video it isn't actually fully rigid.
Not saying its impossible to create a successful system like that, in fact it sounds sort of a natural/organic design that potential could be closer to the optimal design, but it would be more difficult to accomplish and probably need more testing and design iterations. So I think OPs strategy is more simple and pragmatic in this case.
You're going to transfer the energy quickly, you'll just break any "arm" that's light enough to accelerate with the energy available.
You could maybe use an 'already broken' arm (rope, like a whip) but then it's not a trebuchet.
To make things easier, start with an "integrated sling ballistic" (ie: a piece of thread between two lego pieces). Sling release is actually fairly complicated so chop it out!
The arm is generally a rigid lever, and the sling acts exactly as your "gradually reducing mass". In his design you can see the kindof bike gear shape at the axle and eventually coming like an "A" towards the tip (and reducing to "infinite thin-ness" via the string/sling).
After you've built one simple (small) trebuchet, test to failure! How much weight can you stack on it before a component breaks (bucket? arm? axle? sling? frame?). How heavy of an object can you throw before "everything starts messing up"?
My recommendation for ballistics is mini-marshmallow (desktop), regular marshmallow, mega-marshmallow, then switch to "hard" ballistics, eg: mini candy pumpkin (the connoisseurs choice), tennis ball, basket ball, etc.
Once you start throwing dense, solid objects (marbles, golf balls, baseballs) your risk of property damage and injury goes up fairly exponentially, so stick to marshmallows and tennis balls until you feel more comfortable.
Spend lots of time figuring out safe/reliable trigger mechanisms, WITH A SAFETY! For my tennis-ball treb, I did three eyebolts with a dumb screwdriver on a string, and a vise-grip (clamping) on the far side. Eventually, you're trying to safely "release" 100's of pounds that is being held in tension, and if you have a misfire (fail to release trigger) you're potentially approaching a "loaded" machine which is demonstrably "not working right". Lifting/loading 100's of pounds and then having an unlocked trigger (which _also_ may fail) is not safe for anyone.
Aerodynamic drag on the kinds of things that are involved in trebuchets is always quadratic with velocity.
Had a similar feeling come to me as I watched this video a few weeks ago. It made me really want to see someone make a whip-based trebuchet that uses a big spring like https://www.youtube.com/watch?v=PkQZU-CBwhk
Either way, the weight of the whip is hardly important right, as you are just doing an energy transfer from the falling weight into the small ball, the lighter the whip the better, as long as it doesn't break. If the whip is heavier, it will soak up the energy budget in momentum
The big issue in your proposed version is building an arm that is light enough to not absorb a lot of energy (it still needs to be rotating at the same speed as the progressive version in the video) but rigid enough to survive that sudden acceleration. My bet is you'd lose way more to the arm flexing and the added weight than you would the benefit of not dealing with aerodynamic losses.