Autonomous excavator constructs a six-metre-high dry-stone wall | Hacker News Reader

Autonomous excavator constructs a six-metre-high dry-stone wall

ethz.ch

568 pointsbelter185 comments

Having made a few dry stone walls I can attest to the merit of this achievement. Judging where rocks are to be placed, and in what orientation is a zen art.

Anecdotally... I once tried to get wild flowers to grow on a stone wall. Every year we would seed the wall, only for the seeds to fall out or be eaten by birds. On inspiration, my boss filled a few shotgun cartridges up with seeds and fired them at the wall. Perfect result! The following year it was covered in flowers.

Now this is gardening I can get behind

Well please don’t stand in front of it…

Not only is this a technical marvel, but it has potential construction physics implications.

In simple terms, currently we expend energy to fragment rock, and then we expend additional energy (as concrete) to glue it back together. We do all this to re-arrange rocks into a shape (and strength) we want, and most importantly to do it cheap.

Suddenly, an alternative just got a lot cheaper.

Clearly this isn't fully baked yet (and it's hardly a drop-in replacement for reinforced/pretensioned/cantilevered/etc) but for things like heavy infrastructure and civil engineering this might offer a way to achieve fundamental thermodynamic improvements.

Excited to see where this technology goes.

The problem with dry stone construction is that you cannot determine the performance characteristics required for, say, a building foundation. Not all boulders are created equal, but a professional concrete pour, mixed and installed to the manufacturer’s specification is going to have a guaranteed maximum strength- something random boulders will not have.

What we can do is create our own concrete mix using Portland cement (the “glue”) and locally sourced materials to bulk it up such as gravel, aggregates, sand, etc.

So the question becomes, how do existing concrete manufacturers determine whether locally sourced materials ("random rocks") are suitable?

In traditional dry stone construction, strength of each rock is determined by sound and by sight. It seems crazy, but nowadays this should be feasible with a neural net. If you want to get fancy (or achieve high reliability) add ultrasound in the grippers or other sensor modalities.

Or... you target a market other than building foundations? :) I suspect different groups could pursue both paths.

> how do existing concrete manufacturers determine whether locally sourced materials ("random rocks") are suitable?

They test individual samples from each pour on a calibrated press. For large buildings and critical pours, they'll keep & test multiple samples that undergo accelerated aging to make sure they're up to spec for the anticipated lifetime of the building (or until the concrete fully stabilizes over a few months / years).

So could catagorization of local stone based on acoustic and visual indicators, combined with random sampling and strength testing provide the assurance needed that a wall will have the required structural properties?

The point is that the concrete manufacturers pulverize and homogenize the aggregate. This makes random sampling acceptable.

If you haven't pulverized and mixed the inputs, a random sample is merely a random sample. It isn't meaningful unless it comes from a high-entropy source.

So are you saying we couldn’t estimate the structural properties of autonomous built dry walls from local stone? Or that the error bars would be unacceptably large? Or just that this approach would only work with certain stone in certain locations?

I think the problem is more that you cannot at all guarantee a specific performance of the structure ahead of time, because you can't know what the properties of the local rocks will be before you are on site and have measured them all.

Geotechnical engineering is all about taking samples and using wide enough safety margins. And in the vast majority of cases, the end result is a building that doesn't fall down.

Sure, you might have been really unlucky that all your test samples of soil and rock were really strong granite, but the exact spot you put the main structural pile, unknown to you, was all mud and silt... But the chances are so vanishingly unlikely we accept it.

Doesn't that apply to wood too? Yet we use wood to build houses almost everywhere.

People rarely build entirely using found wood on site, at the very least each tree to be used gets selected before you bother felling it. But the overwhelmingly most common use of wood is in the form of processed product: using quality controlled lumber of standardised size.

The interesting thing here is you can feasibly measure them all.

Not ahead of time, but as you build.

Sure, but then it's somewhat up to luck precisely what properties you get. This variation might me manageable if you design enough redundancy into the design, but by necessity you will need more margins than if you used a more controlled material.

A Deep Learning Based Method for the Non-Destructive Measuring of Rock Strength through Hammering Sound Appl. Sci. 2019, 9(17), 3484; https://doi.org/10.3390/app9173484

Acoustic inverse problems are solved routinely in seismology and exploration geophysics. Seeing trough a rock the cracks and internal structure using acoustics seems solvable.

Why not do a stones-wall like in the article, but add cement as glue between the rocks?

The wall will then dam up water or crack instead of shifting slightly, I read in another comment.

I did this for a landscaping project because I was not confident enough that it will stay ..lasted ten plus years and then it started cracking. I am now left with unsightly gray blotches and worried the rest might tumble..

It was/is gorgeous..I did three sections ..mostly for three very large kidney shaped raised beds..moss has grown over the stone and I also tried sticking thyme and mint in some of the suitable gaps..

I guess it’s diff if it’s enclosed and filled with soil. I wish i had the confidence and skill to attempt an actual wall.

I used to drive past a house here where the guy took like three years to complete a 80 ft long dry wall and he planted all kinds of beautiful flowering plants ..it was stunning and when it was done, the flowers and fruit trees popped up with their colours too ..it took forever but he didn’t care..you could tell that it was a labour of love and he was really enjoying it.

That is what I assumed when I first started drywalling. However, in the wild, I can't see such a solution working. Unlike the walls of a house, walls bedded on soil move all the time. In the ever-moving domain of the Welsh hillsides, cement glue would crack within a year.

Ideally, drystone walls are composed of two skins. The way the stones are arranged tends these skins to each incline inwards slightly. In this way, the wall as a whole acquires cohesion through physics.

Several reasons… Primarily, because the ground underneath shifts overtime, particularly in area, subject to a freeze and fall cycle. But you also have to solve for water flow - if you build a solid stone wall with concrete, you also need to add drainage on the backside of the wall. With dry, stacked stone, you can often get away without a complex drainage solution

> Not all boulders are created equal, but a professional concrete pour, mixed and installed to the manufacturer’s specification is going to have a guaranteed maximum strength- something random boulders will not have.

Random boulders surely have some guaranteed maximum strength. Now minimum strength is the real problem.

Modern concrete only lasts a hundred years or so. Stone walls last much longer when correctly constructed.

Presumably the placement algorithm understands compression strength of the local rocks, soil heaving, etc.

So, software addresses the initial problems you describe at construction, and geology/materials solve the problem of needing to inspect it every decade or so, and replace it in 50-100 years.

With concrete you have outsourced, not solved, that problem

There's a lot of drystone walls here, in Aotearoa, but not of buildings.

There is a related technique for buildings, building from rubble, using lime morter.

I live in a building partly built that way

It’s more difficult but still possible. Mountain passes in Switzerland are often built on dry walls

Tensile strength will be a problem. Compression will be solved eventually, the tetris part also. But if there are to be tensile or twisting loads - there will be trouble.

Add steel/fiberglass tension rods and you can trade off compressive for tensile strength.

So... reinforced concrete.

> Not only is this a technical marvel, but it has potential construction physics implications.

I don't think this is true. Gabion walls are already widely used, particularly in geotechnical structures, and masonry structures have been around for literally thousands of years.

The reason modern building technologies such as concrete/steel/composite structures replaced these techniques is that they are far safer, easier to maintain, far more space-efficient and and impose far fewer architectural constraints.

Automating this process helps drive down the cost, but cost is one of many factors and not a critical one.

> Suddenly, an alternative just got a lot cheaper.

Not really. This article just showcases a way to automate the construction of a particular wall type. This by no means implies these walls are now competitive.

Other than aesthetics, I don't see how it's possible to even conceive standard techniques such as gabion walls being replaced by dry-stone walls. The cost of building a gabion wall is literally the cost of a metal mesh and filling it with rocks. Gabion walls are even used by armed forces as rapid-deployment defensive walls in the form of Hesco bastions.

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

I'm surprised the seeds would survive the heat and pressure of a shotgun cartridge blast that contained enough gun powder to embed them in a wall

Seeds are incredibly resilient in many cases. Also, shotgun shells come in a range of power level and configuration. Something like birdshot replaced with seeds would make sense to me.

it didn't embed them into the wall but in cracks and spaces where they could germinate.

It sounds like judging where rocks are to be placed is a simple optimization algorithm

Didn't they do that in The Grinch?

For future walls, try a Cornish hedge. Despite the name, these are actually dry stone walls with an earth core. Mature ones are normally covered in wildflowers. Here's how they're made. http://www.cornishhedges.co.uk/PDF/building.pdf

This isn't really a proper wall though, it's a single thickness stack. I don't think it could handle any lateral force at all without toppling. If this machine could build a double skin wall with batter, wall heads, and hearting throughout the center, I'd be really impressed!

I suspect all those things are just fairly minor software changes...

This is the way you use a gun. What a flowery blossoming place this world would be. Reminds me of https://en.wikipedia.org/wiki/Swords_to_ploughshares

> my boss filled a few shotgun cartridges up with seeds and fired them at the wall.

A friend of mind had a hill in his yard, and hired a guy that knew the local seed mix to hydroseed it. It went well.

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

Can't figure a good word would be for what your boss did. It's not quite pyroseeding...

Ballistic broadcasting?

Nevermind, I change my answer to grapeshotting

Just wanted to share that in my country (Spain) and more concretely in my area (rural part of Catalonia) we have a building technique called “Pedra Seca” literal translation is “Dry Stone”. It basically involves building stuff with just rocks, nothing more. It’s still in use mostly in agricultural fields.

Here are a few examples:

https://dd9de8c7b7.cbaul-cdnwnd.com/b5fbcf0e77b7b72ef95acd62...

https://static1.ara.cat/clip/98e3619e-8dd2-4c0b-828c-254a62e...

The google search:

https://www.google.com/search?client=safari&sca_esv=58408328...

Dry stone walling is also very common in the UK. It's a complex and now expensive skill, but a properly-built (this bit is key) wall will stay put for well over a hundred years because the rocks can shift a little rather then cracking the structure as temperatures change and water drains right out of them.

It's also very locally distinctive because the shape of the stones and therefore the wall layout and texture depends on the stones you naturally have nearby.

https://en.wikipedia.org/wiki/Dry_stone#/media/File%3ADry_st...

Very common in rural Ireland, and having such a wall repaired or made is either going to cost you nothing or a LOT of money.

As with many of the best things in life, they are either free or very expensive.

Sex? :)

(edit: no, that was not an offer, sorry)

It's the opposite really, because you're trying to get your rocks on, not off.

Since you mention it, let's discuss hardcore https://hintonswaste.co.uk/news/what-is-hardcore/ (SFW)

Completely unrelated, but are you the adhesive_wombat I would listen to on SoundCloud when I was younger?

No, it's just a random adjective-noun pair used last time I restarted my aaccount!

It's very common in Croatia, too.

And Malta: https://www.google.com/search?q=malta+rubble+wall

There they are called rubble walls ("hajt tas-sejjieh" in Maltese).

Seems pretty common in Mediterranean countries.

Thank you, there is a great explainer in those search results: https://environmentalmalta.com/wp-content/uploads/2019/12/IR...

In Switzerland as well. "Mur en pierre sèche".

I recently witnessed something similar dating back to pre-Inca times in the archeological city of Quilmes: https://drive.google.com/drive/folders/1-6DnSeACUPUUskeJJIuB...

We have them in Italy too btw, search for “muretti a secco”.

Interesting! I am living in Catalonia now, but in Barcelona. I haven't seen (or paid attention) to a wall like that. Will pay more attention and try to find some nearby. Thanks!

You won't find them in the city haha. That said, Catalonia is a very beautiful place and I encourage you to visit its rural landscapes, like Montserrat, Rupit and Viladrau (and of course, all of the Pyrenees)! Also in the coast I love Tossa de Mar and Sant Martí d'Empúries, near L'Escala.

I rode through a huge area filled with these sorts of walls somewhere wayyy down a dirt road in rural central Mexico, sorta near Zacatecas. It looked like it was once a vineyard, or something like that...I'd never seen walls that looked this way before. Really beautiful. I had no idea that it was originally a European thing.

I'm from Mallorca and this technique is also alive there. We call these "marjades" for the terraces and the walls themselves are called "marges".

https://ca.wikipedia.org/wiki/Marge

This seems like game changing technology. There are a heap of such walls around here, some built long ago by Dalmation immigrants (and some allegedly by German POWs). It's quite labour intensive and highly skilled to create them.

We have some on our property, used to retain the hillside after land slips. They were built with an excavator by a gnarly old guy who was unbelievably skilled.

One thing I wonder is whether these are safe. These things terrify the shit out of me, as one toppling rock of that size could instantly kill a child.

Walls built by human operators of sufficient skill are safe (I know, I went around and tried to wobble every single rock to make sure). But are these built to the same level?

You could potentially shotcrete (spray it with concrete to reinforce) the wall after the robotic excavator is done if you’re concerned about stability. I can’t speak to whether you can highly automate this as well, but it seems possible with current state robotics with human operators.

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

https://www.geostabilization.com/technology/shotcrete-robot/

https://dawinvehicles.en.made-in-china.com/product/jBMnaOKdL...

The ability of a dry stack retaining wall to shift and breathe is a feature, not a bug. For stone walls that require added reinforcement against lateral loads, look into tiebacks or geogrid reinforcement.

Appreciate this. Where can we read more on this engineering? All citations are welcome, I am fascinated by this engineering domain.

There are Civil Engineers here (I only studied it at Uni - I ended up in IT!)

When you deploy a structure like a retaining wall you want to try and ensure that the materials you use will retain their properties as long as possible and if part of it fails, it should not case the whole structure to fail. Add additional design requirements as you like eg colour and texture but always think about function first or you will regret it later!

In a garden setting, you will want to consider: gabions, drystone walling and "sleepers" (large lumps of wood - like railway sleepers).

Some quick material thoughts: Wood is prone to rotting, so ensure it is treated and well drained. Drystone walling can be prone to collapse unless it is allowed to drain properly and plants/weeds should be removed. It should be slanted at around 5 degrees from vertical to resist collapse. Gabions made of galvanized steel wire are extremely strong and resistant to pretty much anything. Devon Popples are an ideal filling for gabions and make a phenomenal structure.

I built a deck part way down in my garden. It is about 5m wide and sticks out about 1.8m. Behind it is a sleeper retaining wall which is about 1.8m high. I angled it back by about 2 degrees from vertical. I laid it on a concrete strip to spread the load and gravel base and back filled with quite a lot of scrap brick and rubbish for drainage. I used some 2m x 10mm stainless steel threaded rods embedded into the conc base to ensure horizontal stability (horizontal shear). I used 180mm, No. 10 passivated screws to keep the wall together whilst I built it and back filled.

"Where can we read more on this engineering"

It's everywhere but you will have to deal with local conditions. I am not convinced you are what you claim.

Another quick tip: resist the urge to stack your stones long-ways parallel to the course of wall. Instead, put the longest dimension perpendicular to the wall.

This makes it much harder for the wall to fall outward as it shifts.

The 5 rules of dry stone:

https://thestonetrust.org/5rules/

Perhaps, and there are lots of different types of stone and how to stack them.

My garden walls are of blue lias which is a bit like slate (https://en.wikipedia.org/wiki/Blue_Lias) it lends itself to flat 1-6" thick blocks that lay into quite neat horizontal rows whilst still being disjoint and allowing water to flow out. Weeds do grow in the cracks and the woody ones at least should be removed.

As you say a thicker wall will be more stable. Water and soil pressure will be most intense at the base, so make the base thicker and thin out as you move up. A slight deviation from vertical will allow for the soil expanding. For example my walls were rebuilt after a dry spell so a few degrees off vertical allows for the clay to absorb water and expand.

> I am not convinced you are what you claim.

Knowing where to look without colleagues or steeping in the field is difficult. When teaching people to code, I often run into the same issue: assuming people are good at googling, or know what Stack Overflow is.

I am very good at googling and other online resource discovery, but that is no replacement for a knowledgable practitioner pointing you towards known good, high quality online resources. You don’t know what you don’t know. "Unknown unknowns."

I will even pay subject matter experts when necessary to bootstrap the research and autodidact process. There is no speed limit when learning, but rails and direction have value until you have enough foundation in a domain. I will absolutely let someone teach if they’re willing to set me on the right path, one of the reasons this forum is so valuable: highly knowledgeable people willing to bestow knowledge for free. And asking is mostly free (assuming you are polite and receptive).

An ounce of prevention is worth a pound of cure, or in this case, an ounce of experience is worth a pound of research.

Also: Wisdom. It is the difference between knowledge and experience. -Data, Star Trek

I'm not finding info on "Devon Popples" for use in gabions. Was that a typo? If so, curious what the ideal filling is.

This would be discussed in an introduction to soil mechanics and foundation design (Braja Das's textbooks used to be the standard reference in the US when I was a student 15+ years ago) but I'm afraid they're too dry for anyone outside the profession. I'm not aware of an author who can animate civ eng subjects for the adult with the talent of David Macaulay. But I can point you to a few "engineering gems" that might pique your interest if you like this sort of stuff: prestressed and post-tensioned concrete (the work of Freyssinet, see Billington's books), readings from John Ochsendorf's class on historic structures in https://ocw.mit.edu/courses/4-448-analysis-of-historic-struc..., bicycle wheel as prestressed structure (the same principle used in some tension-compression stadium roof structures): http://www-civ.eng.cam.ac.uk/cjb/papers/p20.pdf

One further book that might be of interest (though only tangentially related) is Structures: Or Why Things Don't Fall Down by J. E. Gordon.

Look at the Dry Stone Wall Association of Great Britain, or The Stone Trust in the United States. These organizations still teach traditional methods of dry stone walling which cover a lot of these techniques.

Or make the walls squiggly ~~~ for extra structural integrity

Shotcrete is regularly applied via robot in mining. The problem with shotcrete in open cell designs such as this is overspray and plugging of the large voids.

Gabion baskets are likely a better idea if you have stability problems and want to do a similar kind of automation.

This defeats the purpose a little bit though because the environmental benefit is to _not_ use concrete, as producing it has very high CO2 output.

They use shotcrete a lot here in BC where cliffs are close to the road, but usually when it's soil or more loose material. Any rock structures (natural or man made) are typically left bare.

Perhaps concrete in construction and demolition waste could be used.

https://www.bigrentz.com/blog/construction-waste-statistics

"You could potentially shotcrete"

You could stop water draining through the retaining wall and turn it into a dam, which will not end well.

My back garden/yard has a 40m drop to a lawn. The levels are managed by dry stone walls. Soil here is heavy clay with really heavy clay sub soil, so efficient drainage is imperative.

As well as drystone walls, consider gabions which can form phenomenally tough retaining walls. You can plant on them too rather than having to weed drystone walls. Gabions are much more resistant to plants forcing joints apart than drystone walls.

You would still need humans to erect the gabions though, no? Trying to understand the best optimization use case for this product and best fit final results. Please excuse my knowledge gaps! I’m am trying to fill those gaps.

What problem are you trying to fix?

Reduction in human labor for building these types of structures. The world is getting old fast, but infrastructure is still required, both new and maintenance of what exists today. Also, job assistance that reduces the toll on humans in this industry. This prevents treating the human as a consumable when the automation can be human driven, increasing quality of life and potential career longevity (if desired).

Not replacing humans, to be clear, but helping them build better with less effort. Thank you for indulging me. I hope to eventually leave tech to build things that are needed that are built to last (I too am getting old fast).

You can fill gabions that have been pre-made by using a mini excavator. Or you can do it by hand but that's pretty labor intensive, unless you like wheelbarrows and buckets I would advise against it. Personal experience and all that ;)

The important thing with gabion is to realize that they still need careful planning, anchoring, leveling and possibly a foundation to ensure that they aren't going to shift over time.

> You could stop water draining through the retaining wall and turn it into a dam, which will not end well.

Which is why gravel backfill with a perforated pipe at the base (that drains to day light) is specified by many/most building codes.

So what you're saying is we made a problem for ourselves, and then we hacked on additional cost (and plastic) to fix it?

I'm starting to see the advantages of this dry stone wall technique.

A 'sealed' retaining wall where the stones are interlocked or even glued/mortared together would probably have a higher load capacity than one made dry/unadhered. Or someone could simply like the aesthetic of a more 'filled in' wall, regardless of practical considerations.

Eh, the perforated pipe technique can efficiently collect the water and move it somewhere else entirely. It's nice that dry stone can just let water fall through, but then it just runs out at the base. Which might be fine, but you might still want a french drain anyway.

I don't understand. You don't need to fully coat the wall with shotcrete in order to stabilize the rocks. There could still be holes.

If the wall is going to move a little shotcrete ain't going to stop it.

I’d disagree with this. The general rule in mining is “six inches of shotcrete will hold up the world.”

Under compression. If you tried to stop a dry-stack wall from moving, it'd be under tension.

Not to mention the shotcrete | granite rock binding will shred itself after a year or two of daily heat|cool cycling in the sun - the expansion rates of the two materials don't match.

The shotcrete I have been around was sprayed onto hogwire (already on the rock) and then bolted. It was not just coated onto existing rock as if it was a cake icing.

One of the ideas of this project is, to NOT use concrete, to not have CO2 emmisions.

And depending on the "skill" of the robot and its algorithms, it could be very stable. I know that quite some ancient buildings in south america are put together just with stones put together(with precision and shaping of stones). They survived earthquakes and are still standing. So if enough stones of various shapes are avaiable - you won't need concrete. The weight and friction will hold them together.

Visiting Scotland, I was told that the apparent lack of stability is what keeps the sheep from pushing against the dry stone walls there.

Apocryphal or true? No idea. It's intriguing to think that sheep have a sense for stable construction, and that people have learned to deceive it.

They had no problem escaping through them in Clarkson's Farm.

I really enjoyed that show, but I have no confidence that production assistants weren't gallavanting about at night acting as gremlins.

Clarkson is first and foremost an entertainer, and I'm sure he'd agree with that assessment.

Clarkson is the sort of guy who would have both friend and foe knocking over his walls just to mess with him.

Nonetheless, sheep farming can be incredibly difficult, and sheep love to assert just enough independence to be annoying.

the adjoining farm next to our backgarden had sheep that was obsessed with getting into my backgarden - no idea why. they would jump over the 5ft wall with barb on the top.

When we complained he move the sheep to another field, they escaped that field and entered our back garden through a totaly different way - through our front garden. eating loads of our plants - I can understand the pain Clarkson went through (made up or not)

Not true. A well built fieldstone wall is massive, we estimate around .5T per linear yard when we build one to 4ft. with cope stones. Sheep will not knock it over, but cattle can. For that reason, through-stones are allowed to protrude a few inches on normal pasture wall for sheep, but you should make them flush on walls for cattle because the cows will try to use the protrusions to scratch themselves and could knock the wall over.

A well built wall doesn't appear unstable. One of the tests is whether or not you can see any daylight through the wall, and the wall should easily support the weight of several people leaning against it.

These folks have a spin-off called 'Gravis Robotics', pretty interesting stuff going on. Their promotional video contains technical details as well:

https://www.youtube.com/watch?v=yiTIXAAulzI

https://gravisrobotics.com/

Nice. Heavy equipment with many degrees of freedom is hard for humans to control. Some operators are really good, but not all of them.

A backhoe hooked up to a virtual reality input arm with force feedback was built several decades ago. The operator could feel obstructions, and dig out the dirt around a pipe by feel. Never became a product, though. Probably too early.

Do operators currently need to manually control every movement of the arm? I always assumed that such heavy equipment uses inverse kinematics to help the operator.

It is usually one lever per hydraulic cylinder.

The human brain already automates the kinematics.

There isn't much upside to automation and a lot of downside because of the diverse instructions required.

Often a lever, or joystick motion, per function. And site plans can be input to bulldozers and excavators, and with precision sensing they do cool things like preventing operator from digging below grade. Those 'sticking pipe' in trenches, and those putting in landfill cells rave about it because it avoids rework. The old guys don't need it, but they are aging out of the workforce.

https://www.komatsu.eu/en/Komatsu-Intelligent-Machine-Contro...

Here's a modern piece of heavy equipment, a Ponsse tree harvester, with the operator showing how the machine is used.[1] One big joystick for each hand, and a lot of buttons on each joystick. The operator is guiding the beast, but once it grabs onto a tree, there's some automation to cut the tree into preset lengths. Notice how fluid the movements of the machine are.

It's nowhere near autonomous, but it's several times faster than total manual control would be.

[1] https://www.youtube.com/watch?v=OlvByJeAfv0

> It is usually one lever per hydraulic cylinder.

Not so much these days, but the general idea is correct.

Usually there are joysticks (one per hand) that provide control over one movement per axis. So for example the right hand joystick on a mini-excavator usually controls bucket curl (left/right) and boom movement (forward/back). Left hand does boom up/down and rotate left/right. There's an ISO standard for this UI. Old school tractor-loader-backhoe machines (what people in the US call "a backhoe") had one lever per cylinder, so more complicated to control and more levers. An old-style motor grader is the extreme example of lever confusion[1].

Operation of heavy machines relies (still) on the brain's capability to virtualize the movement of the machine in terms of the body's limbs. That's why you need some time training to become proficient. After a few hours you just think "I want the bucket over there" and your hands make the necessary movements without thinking.

There's already lots of automation available on high-end machines such as use of GPS to control cut/fill operations with bulldozers, and laser-transit-based automated dig depth control for excavators.

[1] https://www.cat.com/en_US/articles/for-owners/the-basics-of-...

There's no IK because sometimes the operator needs to be very specific about where the "elbow" or "shoulder" ends up, to avoid hitting some other object. So they have manual control over everything.

If you're going to automate the kinematics, then you either need a manual override mode (which the operators will be unskilled with since it's seldom-used), or you need sensors and rules for the machine to know where it's not supposed to be (clearances around wires, other structures, etc) and that needs to be better than human operators currently do.

Those are theoretically possible, sure, but a valve-per-cylinder is really stinkin' reliable.

The force feedback would be yet another layer of tetchy stuff to keep running in a really harsh environment.

I've know backhoe operators who strongly preferred older equipment, when its too new the pins are too tight and they have no "slop" to work with. That slop is what they're using for sensory input, if you watch them. With attention and experience that's real time feedback.

The whole machine is the force feedback mechanism. You're not just sitting there in a comfy chair pulling levers, you're on top of a collection of systems all singing their own song in response to the varying loads placed on them.

> The force feedback would be yet another layer of tetchy stuff to keep running in a really harsh environment.

Luckily, hydraulics offer a very easy way to do force feedback: Just measure the post-valve cylinder pressure.

That means you can have all your wiring and sensors in the cab far away from the dirt and rain. Makes force feedback pretty easy and cheap.

ETH Zurich[1, 2] is fairly famous as an engineering school. Somehow in my mind it is linked with computer scientist Edsger Dijkstra[3], but according to the Wiki he was never based there.

[1] https://en.wikipedia.org/wiki/ETH_Zurich [2] https://ethz.ch/en.html [3] https://en.wikipedia.org/wiki/Edsger_W._Dijkstra

Dijkstra was close to Niklaus Wirth, who was based there.

ETHZ is quite famous indeed.

Einstein, among others, studied there [1]

[1] https://ethz.ch/en/the-eth-zurich/global/events/digitaler-ei...

Indeed, 22 Nobel Prizes have been awarded to researchers who have or had a connection with ETHZ.

https://www.science.org/doi/10.1126/scirobotics.abp9758

https://youtu.be/P7wmotyKgXc

Nice achievement!

With upcoming multimodal foundation models, there will be many more robots with increasing capabilities soon. The world of robotics will look very different just in the next 5 years.

Some announcements:

China’s plan: “…China aims to be ready to mass-produce humanoids by 2025.” https://www.therobotreport.com/china-plans-to-mass-produce-h...

Tesla Robot: “…it would be low cost and available to the public sometime between 2025 and 2027. ” https://builtin.com/robotics/tesla-robot

You should take both of those announcements with a huge dollop of salt. Even the article you shared is extremely skeptical of the actual progress of Tesla's robot. And Chinese press releases should really not be read into too much - they are often much more for internal signalling then information.

Either way, there are companies which are much farther along than those - Boston Dynamics has a series of extremely impressive robots. They are still very far away from being a consumer technology. The kinds of physical power required to achieve the movements makes them extremely dangerous to be around on any malfunction or glitch, and they are very heavy machinery. Even animals and other humans, which are far more advanced in terms of their ability to control their movements, sometimes accidentally hurt others (stepping on your toes, turning around and hitting you when they didn't know you're there, etc.). Imagine a several hundred kilo machine stepping on your toes with its full weight just because it mis-estimated your movements.

> Imagine a several hundred kilo machine stepping on your toes with its full weight just because it mis-estimated your movements.

"Specifications. Tesla Bot is planned to measure 5 ft 8 in (173 cm) tall and weigh 125 lb (57 kg)." https://en.wikipedia.org/wiki/Optimus_(robot)

I'm aware that Tesla tends to be over-optimistic in its timeline regarding AI and robotics. I do not endorse the estimates without question and yes, some other companies might have made more progress than they did. The citations were meant as examples of the development in the domain. (And Boston Dynamics doesn't seem to plan a consumer launch of a humanoid robot any time soon, so I didn't cite them.)

What made me more optimistic these days are recent progress in multimodal large foundation models which allow robots to perform much better, esp on manual manipulation. See demo from Sep 2023 here: https://www.youtube.com/watch?v=D2vj0WcvH5c&t=18s&ab_channel...

Oops, I significantly exaggerated the weight, thanks for finding that.

Still, 60kg is no laughing matter.

Also, isn't the fact that companies which are way further along in developing robot capabilities are not thinking about wide-scale commercialization decent evidence for it not being viable yet?

A rumor in some articles I read suggests that Boston Dynamics robots might actually use less AI compared to Tesla Bots. This is attributed to their origin in more traditional robot control systems research. Some of the impressive choreographies showcased in their videos were, at least in part, orchestrated by their engineers. This implies that these robots may not perform as effectively in unpredictable and partial information environments, such as the real world.

In contrast, the technology behind Tesla Bots is rooted in its research into AI for self-driving cars, which is designed to work well in unpredictable real-world situations.

I'm not aware of any public confirmation of this rumor though.

I wonder how many decades before someone hooks this up to a webcam and lets people play minecraft in real life.

As someone who works in insurance, I advise you not to wait

Considering it scans all the rocks, you'd think it would be able to fit the rocks together better. I don't expect a Incan wall, but I think it would be able to construct a tighter fitting wall if it were programmed to care.

It could perhaps if it had all rocks available up front to scan and plan the entire wall with. But it retrieves one rock at a time and can only place on the top row somewhere. Given the geometries are random I wouldn’t expect it to reliably have tight fits given how it’s sampling and placing.

It scans 20-30 (a truckload) before building a section of the wall with them.

Exactly as I do as a human when I build such walls (several of them in my life). It’s a very difficult spatial problem to fit the stones into a pretty, stable structure. No doubt something large memories and detailed spatial search on a computer could help with. That doesn’t seem to have happened in this example, though. That wall is horrendously ugly.

Not only ugly, but unstable and sketchy. I wouldn't go near that wall.

I'm getting some light towers of hanoi vibes

In other words, it's using a greedy algorithm

Even then, if you make one long row, you have hundreds of available spots to place it, with at least 6 sides to decide on which orientation to place the stone. If there are no good spots currently, it could place the stone down next to the pile and go collect another stone until a better spot materialized.

No, I was wrong. It has a collection of 30 or so rocks per section and optimizes their placement in the entire section. I’ll still posit the randomness of the geometry requires more rocks than that to find a smooth fitting, and I imagine human wall makers select rocks up front with similar geometries before even beginning out of thousands of rocks then optimizes their placement during construction in a similar but less preside way than the machine.

If you watch the video closer than I did you will see it plans each stone for optimal stability and placement.

An Incan wall is much "easier" (airqoted qualifier).

There's much effort required to square up stone and shape faces, a lot of drilling, wedges, and precision chisel finishing.

However, once it comes to the stacking it's been reduced to a bricklaying exercise (already solved for robots and posted to HN).

Traditional drywall evolved from field clearing, stacking rocks to get them out of the way and to reduce wind across exposed otherwise rocky potential fields.

Stacking and balancing "as they come" or with relative minimal look ahead | reshuffle queuing is a tougher proposition; this is impressive work.

I'm not saying that it's not impressive. I'm just saying that I feel like considering that it doesn't have the human limitations of it being hard to remember every facet of the rocks, that it could do better job still!

Some Incan walls are regular bricks, but others are just slightly-dressed boulders used for drywall polygonal masonry. They're an interesting compromise - some dressing for flexibility, but not so much that you have to throw half the boulder away.

For comparison, a 15th generation Japanese stonemason master doing it by eye: https://youtu.be/f9wGOXXoCuY

For those looking for a more technical overview of the process, the overview video of the research paper is available here: https://www.youtube.com/watch?v=TxpE5yryCTU

I'm surprised centimeter precision is sufficient. I'd have thought millimeter precision or so would be required. It feels like 1cm could accumulate to a gap or could spell the difference between stable points contacting and not.

IIUC cm precision should be fine if the algorithms looking for stable placement accounts for that cm (and reality id imagine it’d account for less precision than that). If a placement needs sub cm precision the algorithm would just discard that option.

Additionally there shouldn’t be any accumulation in a specific direction. Any accumulated gaps would be accounted for when placing the next rock and rectified to within a cm.

Damn, that must have been fun to debug and test!

Perhaps in hundreds of years they'll be able to deploy AI robots to poorer regions of the world to build basic utilities, mine resources, and even whole cities from nothing. Or have them depoloyed to current cities constantly repairing and optimizing infrastructure.

Factorio style.

If there's one thing that lower income countries have an excess of, it's people to perform labor. There's a reason why you don't see countries (e.g. China) introducing more automation until they rise to a certain income level.

Generally you're right, but also

a) You may not have skilled labor easily available for a niche like constructing a dry stone wall out of giant rocks, and

b) sometimes employing that labor via local firms can result in corruption issues.

This is the kind of equipment that you do not want a rogue AI getting in control of.

Most likely these come with periodic OS updates… which is an attack vector

This is incredible because the vast majority of what it takes to build roads, rail tracks, docks, and such infrastructure is all in the clearing and relandscaping of land. It doesn't have to be perfect in terms of aesthetics or seals, but it has to be very stable and able to take the thin layer of road surface, rails, concrete on top and the weight that they'd carry.

To be able to automate the landscaping is to massively reduce the cost of infrastructure projects in terms of labour and materials, but also likely the time to deliver them too.

With this type of advancement, we may finally be ready to tackle the biggest challenge for domestic robots since domestic robot were envisioned: ironing and folding clothes.

You have forgotten the first task of the day: buttering fresh toast.

I was once on a mountain top composed almost entirely of rocks like these. I'd like to give a sand castle artist this excavator and ask them to turn it into a fairy tale.

Anyone under the age of 20 is a Minecraft artist already

- Wouldn’t this type of wall normally be built using heavy machinery without laborers lifting and placing heavy rocks?

- A heavy machinery operator would no longer be needed, but wouldn’t you need 1 or more people to operate and manage the robot as well as verify the solution?

It’s a really cool direction for construction but I wonder what the net/net cost savings would really be in the near future anyway.

There's probably a lot of savings to be had by lifting safety rules. Cram more machines to do more work on site and finish faster.

I think this is great tech, but maybe not for those reasons. On any safe site, people are already very cautious and working well back from the excavator. Some teams have a protocol where you have to make eye contact with the digger driver before you can go anywhere within the sweep of its arm.

But without the excavator operator, can you not cram more autonomous excavators to get more work done? Seems like sufficiently good sensors would allow them to work in each others safety zones for parallel work.

Probably not as the soil they ig needs to go someplace. That means dump trucks will still be the majority of the site.

I don't like wars, but being able to build a wall out of some broken concrete blocks without it being man handled is a good feature I believe.

Also being able to build during a long period of time (during the night for example) make the job less exhausting.

Also one could think of it as an assistant: put some augmented reality display, and let the machine showing to the operator where to put the rock.

Or help for building records: you keep the operator part, but you record the scans and position and weights during construction phase, so you can eventually run simulation later to know how the wall will change over time etc.

First they are coming for the medival castle builders, then they come for the renaissance painters.

In all seriousness though, that's a pretty nice feat of robotics. The excavator has to work with stones that are not standardized and has to understand how they interact, my intuition is that that is a lot harder than building a regular brick wall.

The photo of the wall in the article is obviously digitally altered, just to the left of the person walking. The rocks change into uniform smooth grey, and it doesn’t match the reflection in the water. Is it supposed to be the digital plan of the wall vs the real thing?

It absolutely matches the reflection in the water - you can see the hard hat of the person and the colour transition.

The change in colour will be the result of a change in source of boulder. The whole point is that they are "recycling".

where were the wireframe mesh rocks sourced from?

The excavator creates a mesh of each stone by scanning it "in hand", and updates the settled positions of them as it builds. The left side of the image is the 3D digital twin model that is maintained by the system, captured from the same perspective to show how it matches the real wall.

Yes, I think it's an artistic interpretation of a digital model of the stones - how the autonomous system sees them.

If you zoom in you can see that it's a polygon like mesh. It's certainly supposed to be how the rocks are represented in the system.

For those both unfamiliar with Menzi Muck walking excavators and having a little boy's love of construction equipment, there are plenty of videos online of non-autonomous skilled operators with fancy moves...

Jaw dropping. There's too much engineering work involved. Interfacing with vehicle and tons of sensors. Software.

Would be best if some entity quickly moves towards commercialization/packaging of this solution.

I wish there were more companies in the US that can look at problem and solve it to a great deal of technical complexity. It seems like we’re too focused on large and shiny objects.

Robotics is expensive. One needs a team of software developer, electronics engineer and mechanical engineer to develop something. 3 people over 3 years with expensive hardware and couple millions $ for MVP needed. How many google ads for an app could one buy for this sum?

That is what public funded universities and research is for. Many claim projects like CERN is a waste of money but this is where many new ideas and tech comes from as a side effect of trying to solve problems that arise while trying to build a bigger and more powerful accelerator.

I have a problem with public funded universities and research in Germany. I really don’t like how professors own companies where everything comes from PhD students’ work. Basically they transfer state/grant money into personal pockets via paid work hours. Though I don’t know how this works in Switzerland. Another day I saw a long list of CERN technologies that I can license from them and they were not for free.

The Fabricate conference has always been great for research like this

I assume (but don't know) that this method of construction would be less viable in more seismically-active parts of the world.

Farmers in New England would have loved this. Miles and miles of stone walls just hidden in forests/woods.

Similarly, envision a solar sintering bot operating in the desert

Imagine if the Spartans had this at the battle of Thermopylae

This is the kind of thing computer science academics dream about

One might call it a... Deus Ex Cavator.

This will revolutionise Yorkshire.

Graham Hancock BTFO

Show it still standing in a year and I'll be more impressed.

The Romans could have used this to keep those pesky Germans out!

now the rocks are more rational rocks

Looks safe.

Stonewaller-3000

Minecraft IRL

Reply on news.ycombinator.com