12
$\begingroup$

This plot point has been bothering me for ages: how do robots distinguish between metals? Many robots in fiction just pick up parts of machinery for self-repair or to make more robots. They don't seem to check what metals they are using or perhaps it has something to do with their senses. Do they somehow scan the materials they get their limbs on? Writers usually gloss over this part to save time, but I like to overthink these details. I'd like to explain this phenomena in more scientific detail. This would flesh out more how autonomous robots operate in my story and others as well.

The assumption in this question is that the robot doesn't have access to specialised equipment to make identification tests on materials. Either the robot comes pre-equipped with sensors that identify metals or it uses its pre-existing senses to figure out what they are made out of. If it's specialised equipment, what is the robot equipped with? If it's pre-existing senses, how does it make use of them?

A robots senses would include: sight, hearing, touch (for advanced models), radio and electromagnetic senses.

How does a robot test materials to know what they are? What's the process like?

$\endgroup$
14
  • 5
    $\begingroup$ You know about Handheld Portable XRF Spectrometers - rigth? Also I recomend change hard science to science based - so far we do not have self repairing cyborgs to write scientific works on their behaviour $\endgroup$
    – MolbOrg
    Sep 13 at 9:49
  • 2
    $\begingroup$ In fiction, AFAIK, robots always pick up manufactured parts for self repair, but use of non-fabricated metals is really a special case. In that case, robots must possess a "stomach" which by some technomagic converts a lump of metal into a shiny new part. $\endgroup$
    – Alexander
    Sep 13 at 16:33
  • $\begingroup$ @Alexander like a small CNC turning-milling lathe? $\endgroup$ Sep 13 at 18:09
  • 1
    $\begingroup$ @JulianaKarasawaSouza rather like nanomachines son, if it so sensitive to composition, there is no way some cnc will make a proper heat treating, surface hardening, coatings etc. shape and materials are just base requirements, we can't ommit other processes which go on top of that, lol, not talking about electronics which ... $\endgroup$
    – MolbOrg
    Sep 13 at 19:11
  • $\begingroup$ @MolbOrg can also be a 3D printer too, if we're thinking about plastics $\endgroup$ Sep 13 at 19:13
19
$\begingroup$

How does a robot test materials to know what they are?

The same way we do today in quality control for manufacturing purposes

With X-Ray Fluorescence to check what the part is made of: this technology is widely used in various applications to check for correct metal composition (don't want anyone installing 304 SS pipes instead of the 316L SS pipes that I ordered, thanks), quality of coating and plating of metal parts, etc.

With other Non-Destructive Testing techniques such as regular radiography and ultrasound: the robot can check for defects in the part such as cracks, warps and inclusions (metallic particulates of different density, air or liquid bubbles...)

With Vision Systems: nowadays it is possible to use vision systems to inspect visually 100% of production, there is even a 3D vision system

With Load Cells: your robot can effectively pick something up and know how much it weighs

With the right programming and database to compare stuff with (which is the actual tricky part for the technology today), it is pretty possible to have your robot picking up the parts and analysing them for further use.

$\endgroup$
3
  • $\begingroup$ @MolbOrg Eh, you're mixing standardization with supplier qualification in the comment - they're two distinct processes. Standardization makes sure that everyone works against the same standards of quality (so if I tell my supplier that I want 316L SS, they know what I want without needing me to send the full specification of the material, for example). It does NOT mean that the material is actually following that standard, that's why we test them or qualify suppliers to reduce testing load $\endgroup$ Sep 13 at 10:37
  • 2
    $\begingroup$ Composition of materials IS important today. I can tell you a few horror stories about spectacular failures that happened because different grades of material than specified were used when my NDA expires... $\endgroup$ Sep 13 at 12:17
  • 1
    $\begingroup$ Load cells are not needed to determine weight. So long as you can measure amperage draw on a servo (or pump for pneumatic limbs) and have an inverse kinematic model , you can execute a series of short motions to determine mass/center of gravity, and moments of inertia with respect to axis within a Cartesian frame. sauce: industrial robot programming and automatic load identification routines. $\endgroup$
    – GOATNine
    Sep 14 at 14:00
10
$\begingroup$

Your robots aim is simply to replicate. The exact form/materials are secondary. Only have access to low-grade cast iron? Right, this one's going to be chunky, heavy and not very mobile. Got a load of tungsten mixed in with high carbon steel? Sweet! Hit the gold mine on that one.

So I've mentioned materials by name there, but your robot doesn't need to know exactly what it's building with: Does it conduct electricity? Yup, cool. I can use it for these bits. Is it magnetic? Nup, Right, gotta find something else for those stators. Boes it take more than X force to bend? Oh well, better find something better for those leg braces. Does it anneal/reharden? Cool! Lets run with what we've got. Don't think of a scientist ordering precision parts with metals pure to 99.99% - think of your neighbours workshop where he grabs a chunk of angle iron from the scrap heap and goes "that'll do" before tack-welding it on.

Metals aren't the hard part of replication. Whatever contains the brains is. I'd hate to see the reliability of an electromechanical brain! IIRC you can use thin film metal oxides as semiconductors but you still need tonnes of precision to manufacture them. There's probably a solution somewhere, and please let it involve lasers!

$\endgroup$
3
  • $\begingroup$ This might work for robo-reproduction to some extent, but for self-repair parts have to fit together. To reproduce a design requires decent level of precision. Imagine a humanoid robot with steel legs and an aluminium skull. Now replace the legs with carbon fibre and the skull with cast iron, same shape, everything strong enough. but top heavy and much less stable $\endgroup$
    – Chris H
    Sep 14 at 9:53
  • $\begingroup$ Adapt! Reform! Survive! Those who pick just one body plan are doomed to fall behind. Reimagine yourself. $\endgroup$
    – sdfgeoff
    Sep 14 at 11:08
  • $\begingroup$ The robot/cyborg doesn't have to use cast iron for it's skull just because cast iron is readily available. It can also change later: use cast iron as a quick fix to repair a bullet hole in it's head? When it next finds something better it can swap it out. Just because humans don't like surgically replacing body parts regularly doesn't mean that a robot/cyborg can't..... $\endgroup$
    – sdfgeoff
    Sep 14 at 11:12
6
$\begingroup$

They use a "suction cup" equipped with:

  • A laser pulse to vaporize some of the material
  • a vacuum suction to convey the vapor
  • a mass spectrometer to analyze the vapor and determine its composition

When they touch the material they determine what it is made of, and the related use.

$\endgroup$
1
  • 1
    $\begingroup$ Sounds like a Dalek ;) $\endgroup$
    – DavidT
    Sep 14 at 1:08
2
$\begingroup$

Assuming you're not asking about highly specific metal grades and alloys, the robot distinguishes metals the same way I do. Steel, copper, tin, lead, zinc, aluminum all have different appearances, feel, scent even. Stiffness, hardness, color, density all are part of it. It's a matter of experience and judgment; if you work with metals enough, you just know.

If resistance welding fails, or solder doesn't stick, for example, the workpiece may be aluminum. Sparks from the grinder indicate carbon content of steels. Discoloration from gray to yellowish to purplish occurs as steel is heated. Zinc combusts. Et c.

edit: it seems i just repeated jamesqf's comment: How can I, using just sight, hearing, and proprioception, tell the difference between various metals? Various steels, cast or wrought iron, copper, brass, aluminium, lead - all easy to distinguish with a bit of experience. Now picking out say a particular aluminium alloy from a number of similar ones could be tricky, but would it really matter that much?

$\endgroup$
1
  • $\begingroup$ The robot may have additional senses you don't that allow it to distinguish. If it needs to do so, it probably will. But the basics are the same. $\endgroup$
    – Mary
    Sep 15 at 3:48
2
$\begingroup$

My robots have the ability to generate electromagnetic fields to sense the saturation of the material it is holding. It has a database with the information of all the magnetic metals known, its hysteresis curves, etc. It has a piece of tensorflow logic that senses the composition of the material it is holding, and chooses whether or not the element could be used for the expected result. Doesn't really have the ability to repair itself though.

$\endgroup$
2
$\begingroup$

Parts database

In addition to the other methods mentioned (and I think X-ray fluorescence is the best, if you can carry the equipment around), it might be possible to identify the metal by first identifying the part.

For example, your robot finds a piece of scrap in a boneyard. It examines the object from every angle and searches through an internal database of common components. Maybe there's even a serial number visible.

"Aha!" thinks the robot, "this is the front engine mount from a 2045 Ford Quasar. I know exactly what grade of steel this is!"

This won't work all the time. Maybe the component is unrecognisable or not in the database, but it's quick, easy, and doesn't require specialised sensors.

$\endgroup$
1
$\begingroup$

Presumably, the robot has a blueprint of itself and so it knows what it is (or at least should be) made of and so knows what materials to seek to make a copy of any of its parts.

As for identifying parts, the robot could make use of a spectrometer, there are a few types and depending on what materials the robot is made of it may have just one spectrometer or several. I believe that this will give the most accurate identification of materials.

Current computer vision software struggles to calculate mass and so might not (I'm guessing here) be able to identify materials just from just observing things. I remember that robotic cars don't like plastic bags as they can't figure out what they are and so can't figure out if they have to avoid them or can drive through them safely.

$\endgroup$
6
  • 1
    $\begingroup$ There are even more technological advancements for having robots identifying and sorting stuff, they just tend to be very... dedicated in their purposes. The narrower the task, the faster they do it $\endgroup$ Sep 13 at 10:01
  • $\begingroup$ @JulianaKarasawaSouza I assume you are referring to robots sorting within a production line? That's a much more controlled environment than what a robot that has to gather its own resource will experience. $\endgroup$
    – Lupus590
    Sep 13 at 13:17
  • $\begingroup$ Not necessarily - it can be waste sorting, for example. As I mentioned, the narrower the task, the faster they do it. So a simple vision system can run to the speed of thousands of units per minute (and they do). A complex system like the one I proposed in my answer, will take far, far longer. $\endgroup$ Sep 13 at 13:59
  • $\begingroup$ How could computer vision software possibly determine mass, any more than human vision can? Can you accurately tell the difference between a balloon, a kid's ball, and a bowling ball just by looking? $\endgroup$
    – jamesqf
    Sep 13 at 23:45
  • $\begingroup$ @jamesqf Humans have contextual knowledge that has thus far been impractical to put into computers. We guess mass from how the object moves and our prior experience of objects. Visual clues also contribute: bowling balls usually have three holes for figures, balloons have a tie-off where the air get's blown in, etc. We could train computers to recognise details like the finger holes of a bowling ball and put that into predicting mass, but I'm not aware of any computer system that does this. $\endgroup$
    – Lupus590
    Sep 18 at 11:55
0
$\begingroup$

Resistance, reluctance, hardness, mechanical rigidity tests comparing deflection under known load to thickness and unsupported span.

I'm interested in whether the power requirements of the x-ray fluorescence test represent any barrier to its use.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.