3 ms·

> tractors + wide implements vs robots In theory there exists a crossover point, right? Tractors scale somewhere like O(x^2) to O(x^3) in costs as the implemen

by hansvm 2y ago

> tractors + wide implements vs robots In theory there exists a crossover point, right? Tractors scale somewhere like O(x^2) to O(x^3) in costs as the implements get wider by a factor of x, depending on exact design decisions, and swarms of robots scale out somewhere like O(x). Tractors have been getting bigger (trying to get more work done for the same amount of human labor), and if that trend continues you'd expect some sort of smaller, swarm solution to be viable (be it on the larger end, small automated tractors, or smaller still where you can take advantage of local sensors and automate something like weeding instead of paying for herbicides and that particular flavor of special seeds).

hedora 2y ago
The optimal size for tractors is more a function of the topology of the land and the soil type than $/acre spent on the equipment. In theory, the minimum feasible unit of equipment should be a lot smaller for sand than dirt (less weight is required to break up the sand), and beaches tend to be curved and bumpy, so that favors a swarm of small robots. However, the cost of maintaining the swarm is linear in swarm size, but the cost of maintaining one tractor is sublinear in tractor size. I'd guess it's roughly as difficult (within a factor of 2-10x) to unclog up the sand sifter on a beach combing tractor as it is to unclog the analogous component on a single swarm robot. At 10x, that suggests a swarm of over ~ ten robots is infeasible (unless they are self-repairing and self-maintaining). The article mentions places where wheeled devices can't reach (stairs, steep beaches). I'm guessing that'll be the main use case for these things. As for precision herbicides + planting, I suspect that's better done with special tractor attachments that have fine grained control over each row's implement. Most stuff is grown in large fields and GPS guided tractors are getting more common.
- Animats 2y ago
  There are small beach cleaning machines, down to push-broom sized devices used for cleaning beach volleyball courts. The beaches that need regular cleaning are usually big, heavily used public beaches, where a large machine is effective.
- hansvm 2y ago
  As far as I can tell, your position is something like: (1) People only use bigger tractors for worse soil, not to reduce their own human workloads. This caps the natural progression of machine sizes I'm describing, thus stopping things before a size where swarms matter (not refuting a crossover point (kindly consider the O(x^2) or greater vs O(x) cost idea), but suggesting we'll never reach it). (2) The human cost of maintenance dwarfs all other costs, and the status quo of small vs large machines suggests large machines will be cheaper to repair, so a swarm of 1-2 big machines will be much cheaper to repair than 100-200 small machines. (*) some addendum addressing my less important side-benefits Does that sound like a correct characterization?
GuB-42 2y ago
I am always skeptical about using small machines if big machines can do the job as it is the case with farming, but also energy production, transport, etc... For tractors, most modern "big" tractors already do most of the work automatically. The guy in the cabin is mostly there for supervision and dealing with the unexpected. And later, for maintenance. Swarms won't help, quite the opposite in fact: more supervision, more unexpected situations, more hardware to maintain... Better automation can help, but better automation will help big machines too. Weeding could potentially be done by an wide array of sensors and picker arms, pulled by a tractor, and maybe followed by a large collector. This way, there is no need to have a propulsion and power system for every picker, the tractor will provide this. Sometimes, small machines make sense but the general trend is towards bigger, it the technology allows it.
- hansvm 2y ago
  Kindly ignore the weeding and automation; that was a side-benefit. The core of the argument was O(x^2) vs O(x) costs. What keeps that from mattering?
  - GuB-42 2y ago
    It is not O(x^2). If you look at this https://www.farmtender.com.au/articles/john-deere-tractors-prices https://www.farmtender.com.au/articles/john-deere-tractors-p... you will see that the price of a new tractor increases more or less linearly per HP. HP should be a decent approximation for the amount of work done. And in many cases, it is sublinear. A semi-truck costs much less per ton than a van. Fright trains are even cheaper. And container ships, even less so. In fact, I am convinced that it is sublinear for tractors too, but the benefits go to John Deere rather than to the customer: bigger tractors offer more value to the customers who have a use for them, so they can increase their prices accordingly.
    - hansvm 2y ago
      > price of a new tractor Working in tech, you should know that the list price of an item is a poor proxy for how much work and how many raw materials went into it. When you don't have very many brand choices, market dynamics matter quite a bit. I think it's noteworthy that those same tractors increase in weight superlinearly with horsepower. Let's assume price is a good enough proxy metric though: > HP should be a decent approximation for the amount of work done. The problem is that the amount of work done, measured in energy at the tractor, isn't a good approximation of the amount of work done, measured in useful effects on the farmland (stated differently, wider implements are less efficient). If you look at what width of tiller or other implement you can pull behind those tractors, it scales poorly with horsepower, and when you read through case studies of people who tried to get away with smaller tractors and wider implements you find that they had a lot more human toil and more passes over the farmland to make them work. A more useful proxy metric would be how many dollars of tractor a single person needs to manage X acres of farmland (for any fixed task) in Y hours. That's a little hard to gather data on because you have qualitative shifts in who's farming for which reasons as you increase acreage, but if you look at the actual tractors' capabilities (e.g., appropriate width of implements) you can compute it as a derived metric. > comparison to trains, semis, ... The comparison isn't great. The whole point of a tractor is to manipulate its environment, immediately spending energy to break physical bonds in plants, accelerating and decelerating a rock from point A to a nearby point B, .... Mechanisms of transport get around that because the only "mandatory" costs are the delta-v: accelerating and decelerating the load. If you're transporting a long distance, that's a negligible fraction of the costs, and all you're doing with scale is reducing the drag along the way. Since drag is nearly proportional to frontal surface area in common speed/weight regimes, you get the observed effect that a longer vehicle (a van or a semi) is more efficient per unit of load than a shorter vehicle (with extra benefits because the shape itself has less drag per unit of frontal surface area). As you scale to something like a train, rolling resistance matters a lot more. The extra efficiencies there are basically because your steel wheels don't deform as much as rubber wheels, and more of that deformation is reclaimable (with many other benefits, e.g., you're able to waste less weight on the engine relative to the load). The key feature making that possible though is transport over long distances, amortizing the cost of startup and shutdown, with a secondary important feature being low rolling resistance pathways (roads, bridges, rails, cargo ships with the barnacles cleaned off, ...). Farms have neither of those. Loose dirt is piss-poor rolling resistance characteristics, and the bulk of the work you're doing is severing physical bonds to move things a few inches or feet. The "work" being done is the whole point of the activity, rather than a byproduct you're able to minimize.

by hansvm 2y ago

hedora 2y ago

The optimal size for tractors is more a function of the topology of the land and the soil type than $/acre spent on the equipment. In theory, the minimum feasible unit of equipment should be a lot smaller for sand than dirt (less weight is required to break up the sand), and beaches tend to be curved and bumpy, so that favors a swarm of small robots. However, the cost of maintaining the swarm is linear in swarm size, but the cost of maintaining one tractor is sublinear in tractor size. I'd guess it's roughly as difficult (within a factor of 2-10x) to unclog up the sand sifter on a beach combing tractor as it is to unclog the analogous component on a single swarm robot. At 10x, that suggests a swarm of over ~ ten robots is infeasible (unless they are self-repairing and self-maintaining). The article mentions places where wheeled devices can't reach (stairs, steep beaches). I'm guessing that'll be the main use case for these things. As for precision herbicides + planting, I suspect that's better done with special tractor attachments that have fine grained control over each row's implement. Most stuff is grown in large fields and GPS guided tractors are getting more common.

Animats 2y ago
There are small beach cleaning machines, down to push-broom sized devices used for cleaning beach volleyball courts. The beaches that need regular cleaning are usually big, heavily used public beaches, where a large machine is effective.
hansvm 2y ago
As far as I can tell, your position is something like: (1) People only use bigger tractors for worse soil, not to reduce their own human workloads. This caps the natural progression of machine sizes I'm describing, thus stopping things before a size where swarms matter (not refuting a crossover point (kindly consider the O(x^2) or greater vs O(x) cost idea), but suggesting we'll never reach it). (2) The human cost of maintenance dwarfs all other costs, and the status quo of small vs large machines suggests large machines will be cheaper to repair, so a swarm of 1-2 big machines will be much cheaper to repair than 100-200 small machines. (*) some addendum addressing my less important side-benefits Does that sound like a correct characterization?

GuB-42 2y ago

I am always skeptical about using small machines if big machines can do the job as it is the case with farming, but also energy production, transport, etc... For tractors, most modern "big" tractors already do most of the work automatically. The guy in the cabin is mostly there for supervision and dealing with the unexpected. And later, for maintenance. Swarms won't help, quite the opposite in fact: more supervision, more unexpected situations, more hardware to maintain... Better automation can help, but better automation will help big machines too. Weeding could potentially be done by an wide array of sensors and picker arms, pulled by a tractor, and maybe followed by a large collector. This way, there is no need to have a propulsion and power system for every picker, the tractor will provide this. Sometimes, small machines make sense but the general trend is towards bigger, it the technology allows it.

hansvm 2y ago
Kindly ignore the weeding and automation; that was a side-benefit. The core of the argument was O(x^2) vs O(x) costs. What keeps that from mattering?
- GuB-42 2y ago
  It is not O(x^2). If you look at this https://www.farmtender.com.au/articles/john-deere-tractors-prices https://www.farmtender.com.au/articles/john-deere-tractors-p... you will see that the price of a new tractor increases more or less linearly per HP. HP should be a decent approximation for the amount of work done. And in many cases, it is sublinear. A semi-truck costs much less per ton than a van. Fright trains are even cheaper. And container ships, even less so. In fact, I am convinced that it is sublinear for tractors too, but the benefits go to John Deere rather than to the customer: bigger tractors offer more value to the customers who have a use for them, so they can increase their prices accordingly.
  - hansvm 2y ago
    > price of a new tractor Working in tech, you should know that the list price of an item is a poor proxy for how much work and how many raw materials went into it. When you don't have very many brand choices, market dynamics matter quite a bit. I think it's noteworthy that those same tractors increase in weight superlinearly with horsepower. Let's assume price is a good enough proxy metric though: > HP should be a decent approximation for the amount of work done. The problem is that the amount of work done, measured in energy at the tractor, isn't a good approximation of the amount of work done, measured in useful effects on the farmland (stated differently, wider implements are less efficient). If you look at what width of tiller or other implement you can pull behind those tractors, it scales poorly with horsepower, and when you read through case studies of people who tried to get away with smaller tractors and wider implements you find that they had a lot more human toil and more passes over the farmland to make them work. A more useful proxy metric would be how many dollars of tractor a single person needs to manage X acres of farmland (for any fixed task) in Y hours. That's a little hard to gather data on because you have qualitative shifts in who's farming for which reasons as you increase acreage, but if you look at the actual tractors' capabilities (e.g., appropriate width of implements) you can compute it as a derived metric. > comparison to trains, semis, ... The comparison isn't great. The whole point of a tractor is to manipulate its environment, immediately spending energy to break physical bonds in plants, accelerating and decelerating a rock from point A to a nearby point B, .... Mechanisms of transport get around that because the only "mandatory" costs are the delta-v: accelerating and decelerating the load. If you're transporting a long distance, that's a negligible fraction of the costs, and all you're doing with scale is reducing the drag along the way. Since drag is nearly proportional to frontal surface area in common speed/weight regimes, you get the observed effect that a longer vehicle (a van or a semi) is more efficient per unit of load than a shorter vehicle (with extra benefits because the shape itself has less drag per unit of frontal surface area). As you scale to something like a train, rolling resistance matters a lot more. The extra efficiencies there are basically because your steel wheels don't deform as much as rubber wheels, and more of that deformation is reclaimable (with many other benefits, e.g., you're able to waste less weight on the engine relative to the load). The key feature making that possible though is transport over long distances, amortizing the cost of startup and shutdown, with a secondary important feature being low rolling resistance pathways (roads, bridges, rails, cargo ships with the barnacles cleaned off, ...). Farms have neither of those. Loose dirt is piss-poor rolling resistance characteristics, and the bulk of the work you're doing is severing physical bonds to move things a few inches or feet. The "work" being done is the whole point of the activity, rather than a byproduct you're able to minimize.