Hacker News Reader

Comment by ukulele | Hacker News Reader

This may be a dumb question, but at what point does a "machine" become a "robot"?

My definition: there is no clear limit, but there is a spectrum of ‘roboticity’ that corresponds to the spectrum of ‘versatility’. The more versatile a machine is, the more a robot it is.

For instance, a blender is not versatile. But a cooking robot can do more things (it blends, but also cooks, mixes, etc.). This is why the cooking robot is more a robot than a blender.

This is similar in industry. You have specialized machines, which can do a single thing. Industrial robots are more versatile because we can program them to achieve different task (e.g. when there is a new model of car). The ultimate robot would be as versatile as a human. This kind of humanoid would have the highest level of ‘roboticity’.

Why would human versatility define the upper limit?

Great question. I think it stems from the origin of robotics itself. The desire to build a machine that is "alive" and get it to obey at all our commands. Sometimes I think that robots are slaves 2.0, and as such, they will have human traits.

> The more versatile a machine is, the more a robot it is. . . . Industrial robots are more versatile because we can program them to achieve different task (e.g. when there is a new model of car).

Agreed. I actually think movies downplay "robot" arms too much, making them look like dumb graspers/movers when in fact they're incredibly versatile not only mechanically, but also programmatically (for example, being able to specify maximum forces, a center of mass, "up", and have it work with all of these parameters as efficiently as possible).

For example, the Kuka:

https://www.youtube.com/watch?v=Vcl6n4LOJRM

https://www.youtube.com/watch?v=ck8y1sl97BY

https://www.youtube.com/watch?v=bxbjZiKAZP4

It seems, in practice, that a robot is what we abstractly imagine, but once implemented in the concrete, it's demystified and everybody goes, "Oh that? That's just a machine".

I didn't know that CPUs do branch prediction until Spectre, I'm work in the steel construction industry so my knowledge of the works of CPUs is limited.

Anyway, I immediately though of this concept you've described: Artificial Intelligence is an abstract imagine, once implemented it's just "branch prediction", or some such.

To be fair, I don’t think anyone working in CPUs ever claimed that branch prediction is AI.

As a matter of fact the two do: [0]. It appears now common place to use some kind of machine learning system in a branch predictor. Which some people consider to be AI.

0 - http://www.electronicdesign.com/microprocessors/ai-helps-amd...

When it senses the item being worked on, and adjusts accordingly.

If it always does exactly the same thing, even if triggered, then I would call it a machine.

So you could have a packing machine - each item to pack is always in the same place. Or a packing robot - it looks for the item to pack, and grabs it where it finds it.

There are obviously some machines/robots that fall on the fuzzy line between those. For example if the machine is simply triggered by the arrival of an item to pack, it's kind of borderline.

Basically: Robot = Machine + Sensors

So closed loop control qualifies a machine as a robot? I'd think it also needs some degree of generality, a robot arm can be reprogrammed for different tasks but something like a VSD cannot.

Yup, generality and reprogrammability would be my criterion as well. As for closed loop control, well maybe? Just about everything meant for industry uses servos, so there's a closed loop for positioning, but it won't necessarily know anything about the work itself.

What's a VSD in this case? Variable Speed Drive?

Yeah, sorry - VSD = variable speed drive. They often have a built-in PID controller which takes an analog input and controls the motor speed accordingly, so for example you can run drive a pump at the required speed to produce a required flow rate.

Yes, I'd say that a the essence of a robot is that it uses sensors to adjust its output actuators to achieve a goal.

The bread and butter of the industrial robotics industry is SCARA machines for pick 'n place. These commonly don't have any sensors beyond a few rotary encoders and motor feedback. They're just moving between preprogrammed points. It's hard to make a simple definition that includes everything.

"These commonly don't have any sensors beyond a few rotary encoders and motor feedback."

you contradicted your own point. Those are sensors. Without them the P&P wouldn't be able to do its job.

I could take the encoders out and the robot would work fine. Its position would just grow increasingly inaccurate over time and make the production line more fragile. It would also make writing the pick n place code a bit of a pain for several reasons, but they're not fundamental to the design.

I mean... the robot would decidedly not work fine without joint encoders, on any timescale relevant to the task. You likely could not accomplish one pick/place operation. You need the position feedback because you don't have a good enough model of the motors (and everything) to do feedforward control alone. So I'd say position feedback is pretty fundamental to the design. (If you have stepper motors, then okay, you do have a good enough model of the motors.)

I think I agree with your point at large, though, that the "proprioceptive" sensors of the robot don't have anything to do with the environment the robot is in, or the task at hand.

But now what about an air conditioner with a thermostat? That's got a sensor and an actuator and the sensor in this case really is fundamental to the task at hand.

a thermostat, under my definition would indeed be a robot- although a fairly boring one since it doesn't move itself, just modulate room temperature.

A toilet cistern would also qualify. (And in fact, while I wouldn't deem it a robot, I often use this as the simplest possible control system example.)

nope, encoders are definitely fundamental. You cannot give a DC (brushed or not) motor voltage and have it turn a fixed distance. The closest thing is steppers, but once they stall, you've lost feedback and positioning information.

Here’s an entire podcast of short episodes that started out dedicated to this question. They’re now branched out to answering questions about definitions in general.

https://www.theincomparable.com/robot/

When does a group of grains turn into a heap?

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

People can list any number of things: reactive behaviour, complexity of tasks, programmability, autonomy, anthropomorphic design (in part or whole). But really, whether we call it a robot or a generic machine depends on which probabilistic impulses any given instance trigger on the average person using the language. The criteria differ from person to person, the activation levels differ too, and the probability distribution can change over time also. The word is a cultural artifact and has no real meaning other than its relations with concepts in people's brains. Unlike, say, the word "two": that has very precise meaning independent of cultural concepts.

I think the real definition, which the author explicitly denies is "roughly human shaped". I assume that's why they referred to "arms" and "seeing" in an attempt to make their machines more robotic.

As a fan of the "How It's Made" show, I'm fascinated by the systems in factories, which could be thought of as large machines themselves, even if some of the cogs are human beings doing repetitive tasks.

Apparently though, people don't care about jobs being "stolen" by fork lift trucks or giant ovens but human shaped robots let your job get taken without even the smallest amount of rijigging and that makes the threat very real for some people.

When it senses and responds to an unstructured environment, such as autonomous cars. If you can drop the device into an unstructured environment and it will be able to do something useful, it's a robot. If it operates in a fixed, defined environment, particularly if it's repetitious, it's automation. I'd guess the cheese robot is actually automation, by my definition.

I use a very simple definition. All robots are machines, but not all machines are robots. A robot is an autonomous mechanical worker. That is, barring exceptional circumstances (e.g. less than 16 strings of cheese), something goes in the input and gets transformed in some good at the output without human intervention. A machine that replaces human labour.

> A machine that replaces human labour.

We also need a word for a human that replaces machine labour. E.g. in the case of downtime, or when there is no funding to buy one.

Perhaps "Mechanical Turk" would be that word. So that would mean that "robot" can be defined as its opposite.

I think enterprise / business terminology is heading towards "robot" is what replaces a person doing an existing job (which is why i am surprised to find the moral equivalent of perl scripts being called robots when reporting at board level) and what we might more commonly call AI is for as yet untried opportunities.

but that might just be a limited pov

I'd say a machine is a robot if, and only if, it demonstrates autonomy and mobility (either by locomotion of the whole apparatus, or by motion of a member thereof).

Mobility is a classifier that helps distinguish robots from computers. Autonomy differentiates robots from piloted vehicles.

Autonomy does not require inference, reasoning, or behavior choice.

I'd say a robot is "a machine capable of carrying out a complex series of actions automatically, especially one programmable by a computer."

uhm, that would include a cnc lathe

So? Some of the high end ones, especially the one with arms for automated tool switching and stuff, I would definitely call robots.

Which replaced jigs, which replaced master machinists.

Master machinists weren’t replaced, they learned how to write G code. The underlying fundemental knowledge is still extremely valuable.

A fully automated CNC lathe with feedback is a robot.

For me personally (I'm a hobbyist roboticist), when you put in sensors and use those for a closed control feedback loop to achieve an objective.

Perhaps when it can carry out programmable tasks.

Reply on news.ycombinator.com