When doing robot control, how do you model in the control of the robot? Do you have tool_use / function calling at the top level model which then gets turned into motion control parameters via inverse kinematic controllers?
What is the interface from the top level to the motors?
I feel it can not just be a neural network all the way down, right?
Have a look at the post - it explains how it works. There are two models: a 7-9Hz 7B vision-language model, and a 200Hz 80M visuomotor model. The former produces a latent vector, which is then interpreted by the latter to drive the motors.
> a 7-9Hz 7B vision-language model, and a 200Hz 80M visuomotor model.
huh. An interesting approach. I wonder if something like this can be used for other things as well, like "computer use" with the same concept of a "large" model handling the goals, and a "small" model handling clicking and stuff, at much higher rates, useful for games and things like that.
This is typical in real time applications. A supervisor tries to guess in which region the system is currently and then invokes the correct set of lower level algorithms.
You don't use function calling. You specifically train the neural network to directly encode the robot action as a token. There are many ways. You can output absolute positions, delta positions, relative trajectory. You can do this in joint space or end effector space.
200Hz is barely enough to control a motor, but it is good enough to send a reference signal to a motor controller. Usually what is done is that you have a neural network to learn complex high level behaviour and use that to produce a high level trajectory, then you have a whole body robot controller based on quadratic programming that does things like balancing, maintaining contacts when holding objects or pressing against things. This requires a model of the robot dynamics so that you know the relationship between torques and acceleration. Then after that you will need a motor controller that accepts reference acceleration/torque, velocity and position commands which then is turned into 10kHz to 100kHz pulse width modulated signals by the motor controller. The motor controller itself is driving MOSFETs so it can only turn them on or off, unless you are using expensive sinusoidal drivers.
"The first time you've seen these objects" is a weird thing to say. One presumes that this is already in their training set, and that these models aren't storing a huge amount of data in their context, so what does that even mean?
It probably gives them confidence that they can accurately see a thing even though they don't know what that thing is.
I could also imagine a lot of safety around leaving things outside of the current task alone so you might have to bend over backwards to get new objects worked on.
These models are trained such that the given conditions (the visual input and the text prompt) will be continued with a desirable continuation (motor function over time).
The only dimension accuracy can apply to is desirability.
So from what I understand it actually means that they were for example never trained on a video of an apple. Maybe only on a video of bread, pineapple, chocolate.
However, as it was trained using generic text data similarly to a normal LLM, it knows how an apple is supposed to look like.
Similar than a kid that never saw a banana, but his parent described it to him.
Yeah, except that future doesn't need us. By us I mean those of us who don't have $1B to their name.
Do you really expect the oligarchs to put up with the environmental degradation of 8 billion humans when they can have a pristine planet to themselves with their whims served by the AI and these robots?
I fully anticipate that when these things mature enough we'll see an "accidental" pandemic sweep and kill off 90% of us. At least 90%.
I'd expect Musk and Bezos to know about von Neumann replicators; factories that make these robots staffed entirely by these robots all the way to the mines digging minerals out of the ground… rapid and literally exponential growth until they hit whatever the limiting factor is, but they've both got big orbial rockets now, so the limit isn't necessarily 6e24 kg.
To me this is such a weird wish. Why would you not want to care for your home and the people living there? Why would you want to have a slave taking these activities from you?
I'd rather have less waged labour and more time for chores with the family.
It's called a "house cleaner" and they only cost ~$150 (area and all varies) bi-weekly. I'll shit a brick (and then have the robot clean it up) if a robot is ever cheaper than ~$4000/yr.
Why not? If it's done with all the chores, I can have it make some silly woodworking / art project for Etsy to earn its keep, or just loan it out to neighbors.
Well perhaps not at night, but otherwise there’s always something to clean, something to fix, something to cook, take care of the yard.. heck I might need two robots ^^’
I'd pay $2k for something that folds my laundry reliably. It doesn't need arms or legs, just like my dishwasher doesn't need arms or legs. It just needs to let me dump in a load of clean laundry and output stacks of neatly folded or hung clothing.
With the way they move, they look like stoned teenagers interning at a Bond villain factory. Not to knock the tech but they're scary and silly at the same time.
Well, when you put it like that I feel a bit uncomfortable...
But it did seem like title of their mood board must have been "Black Mirror".
Very uncanny valley, the glossy facelessness. It somehow looks neither purely utilitarian/industrial nor 'friendly'. I could see it being based on the aesthetic of laptops and phones, i.e. consumer tech, but the effect is so different when transposed onto a very humanoid form.
Until we get robots with really good hands, something I'd love in the interim is a system that uses _me_ as the hands. When it's time to put groceries away, I don't want to have to think about how to organize everything. Just figure out which grocery items I have, what storage I have available, come up with an optimized organization solution, then tell me where to put things, one at a time. I'm cautiously optimistic this will be doable in the near term with a combination of AR and AI.
Maybe I don't understand exactly what you're describing but why would anyone pay for this? When I bring home the shopping I just... chuck stuff in the cupboards. I already know where it all goes. Maybe you can explain more?
Maybe some people just assume there is a “best” or “optimal” way to do everything and AI will tell us what that is. Some things are just preference and I don’t mind the tiny amount of energy that goes into doing small things the way I like.
One use case I imagine is skilled workmanship. For example, putting on a pair of AR glasses and having the equivalent of an experienced plumber telling me exactly where to look for that leak and how to fix it. Or how to replace my brake pads or install a new kitchen sink.
When I hire a plumber or a mechanic or an electrician, I'm not just paying for muscle. Most of the value these professionals bring is experience and understanding. If a video-capable AI model is able to assume that experience, then either I can do the job myself or hire some 20 year old kid at roughly minimum wage. If capabilities like this come about, it will be very disruptive, for better and for worse.
Have you met people that seem to be able to fix almost anything?
If you can't get a tutorial on your exact case you learn about the problem domain and intuit from there. Usually it works out if you're careful, unlike software.
Then you are equally fucked as the AI will be, so no difference.
Case in point, I remember about ten years ago our washing machine started making noise from the drum bearing. Found a Youtube tutorial for bearing replacement on the exact same model, but 3 years older. Followed it just fine until it was time to split the drum. Then it turned out that in the newer units like mine, some rent-seeking MBA fuckers had decided more profits could be had if they plastic welded shut the entire drum assembly. Which was then a $300 replacement part for a $400 machine.
An AI doesn't help with this type of shit. It can't know the unknown.
But once it knows it’s pretty certain to become common knowledge almost instantaneously. That’s not possible now. What you learn stays localised to you and may be people 1 degree away from you that’s it.
How does that work? None of the current AI models can re-train on the fly. How would the inference engine even know if it's a case of new information that needs to be fed back, or just a user that's not following instructions correctly?
This is correct. What I meant to say was that in due course, re-training on the fly will become a norm. Even without on the fly re-training we are looking at a small delta.
It would be nice to be able to select a recipe and have it populate your shopping list based on what is currently in your cupboards. If you just chuck stuff in the cupboards then you have to be home to know what they contain.
Or you could wear it while you cook and it could give you nutrition information for whatever it is you cooked. Armed with that it could make recommendations about what nutrients you're likely deficient in based on your recent meals and suggest recipes to remedy the gap--recipes based on what it knows is already in the cupboard.
I took home ec in 2001. I learned to use a sewing machine, it was great.
But none of the kitchen stuff we learned had anything to do with ensuring that this week's shopping list ensures that you'll get enough zinc next week, or the kind of prep that uses the other half of yesterday's cauliflower in tomorrow's dinner so that it doesn't go bad.
These aren't hard problems to solve if you've got time to plan, but they are hard to solve if you are currently at the grocery store and can't remember that you've got a half a cauliflower that needs an associated recipe.
Presumably, they won't as this is still a tech demo. One can take this simple demonstration and think about some future use cases that aren't too different. How far away is something that'll do the dishes, cook a meal, or fold the laundry, etc? That's a very different value prop, and one that might attract a few buyers.
The person you're replying to is referring to the GP. The GP asks for an AI that tells them where to put their shopping. Why would anyone pay for THAT? Since we already know where everything goes without needing an AI to tell us. An AI isn't going to speed that up.
You want to outsource thinking to a computer system and keep manual labor? You do you, but I want the opposite. I want to decide what goes where but have a robot actually put the stuff there.
That's the problem, though - the computer is already better at thinking than you, but we still don't know how to make it good at arbitrary labor requiring a mix of precision and power, something humans find natural.
In other words: I'm sorry, but that's how reality turned out. Robots are better at thinking, humans better at laboring. Why fight against nature?
I think he means outsourcing everything eventually, but right now, outsourcing the thought process is possible, while outsourcing the manual labor is not.
A good AI fridge would be already a great starting point. With a checkin procedure that makes sure to actually know whats in the fridge. Complete with expiry tracking and recipe suggestions based on personal preferences combined with product expiry. I am totally unimpressed with almost everything I see in home automation these days, but I'd immediately buy the AI fridge if it really worked smoothly.
I imagine a something like a headlamp except it's a projector and a camera so it can just light up where it wants you to pick something up in one color or where it wants you to put it down in another color. It can learn from what it sees of my hands how the eventual robot should handle the space (e.g. not putting heavy things on top of fragile things and such).
I'd totally use that to clean my garage so that later I can ask it where the heck I put the thing or ask it if I already have something before I buy one...
I fully agree, building something like this is somewhere in my back log.
I think the key point why this "reverse cyborg" idea is not as dystopian as, say, being a worker drone in a large warehouse where the AI does not let you go to the toilet is that the AI is under your own control, so you decide on the high level goal "sort the stuff away", the AI does the intermediate planning and you do the execution.
We already have systems like that, every time you use you tell your navi where you want to go, it plans the route and gives you primitive commands like "on the next intersection, turn right", so why not have those for cooking, doing the laundry, etc.?
Heck, even a paper calendar is already kinda this, as in separating the planning phase from the execution phase.
I'm quite slowly working on something like this, but for time.
For "stuff" I think a bigger draw is having it so it can let me know "hey you already have 3 of those spices at locations x, y, and z, so don't get another" or "hey you won't be able to fit that in your freezer"
Yeah there’s more to it than that. Do you want a can of beans to be put in the utensil draw just because it would fit? If it was done as you describe the placement of all of your items would be almost random each time, the bot need to have contextual memory and familiarity with your storage habits and preferences.
This can be done of course, in your statement the phrase “just figure out” is doing a lot more heavy lifting than you allude to
Dunno, I would not want to bleed my mental faculties for doing even simple planning work like this by outsourcing it to AI. Reliance on crutches like this would seem like a pathway to early-onset dementia.
This is almost literally the first chapter in Marshall Brain's "Manna" [0], being the first step towards world-controlling AGI:
> Manna told employees what to do simply by talking to them. Employees each put on a headset when they punched in. Manna had a voice synthesizer, and with its synthesized voice Manna told everyone exactly what to do through their headsets. Constantly. Manna micro-managed minimum wage employees to create perfect performance.
Imo, the Terminator movies would have been scarier if they moved like these guys - slow, careful, deliberate and measured but unstoppable. There's something uncanny about this.
If you're training robots to interact with humans this is the kind of behavior you'd want. Humans use a ton of nonverbal hints like this to register context.
So, there's no way you can have fully actuated control of every finger joint with just 35 degrees of freedom. Which is very reasonable! Humans can't individually control each of our finger joints either. But I'm curious how their hand setups work, which parts are actuated and which are compliant. In the videos I'm not seeing any in-hand manipulation other than just grasping, releasing, and maintaining the orientation of the object relative to the hand and I'm curious how much it can do / they plan to have it be able to do. Do they have any plans to try to mimic OpenAI's one handed rubics cube demo?
Telling a robot verbally "Put the cup on the counter" and having it figure out what the cup is, what the counter is in its field of view would have seemed like science fiction. The object manipulation itself is still well behind what we saw in the 2015 DARPA Grand Challenge though.
Does anyone know how long they have been at this? Is this mainly a reimplementation of the physical intelligence paper + the dual size/freq + the cooperative part?
They appear to be realtime, based on the robot's movements with the human in the scene. If you believe the article, it's zero shot (no preprogramming, practice or special training).
I get the impression there’s a language model sending high level commands to a control model? I wonder when we can have one multimodal model that controls everything.
The latest models seemed to be fluidly tied in with generating voice; even singing and laughing.
It seems like it would be possible to train a multimodal that can do that with low level actuator commands.
If you read the article, they describe a two-system approach; one "think fast" 80M parameter model running at 200hz to control motion, and one "think slow" 7B parameter model running at ~7-9hz for everything else (scene understanding, language processing, etc).
If that sounds like a cheat, neuroscientists tell us this is how the human brain works.
Wonder what their vision stack is like. Depth via sensors or purely visual and the distance estimating of objects and inverse kinematics/proprioception, anyway it looks impressive.
I don't know, there has been so many overhyped and faked demos in humanoid robotics space over the last couple years, it is difficult to believe what is clearly a demo release for shareholders. Would love to see some demonstration in a less controlled environment.
Imagine they bring one out to a construction site and they treat the robot as a new rookie guy, go pick up those pipes. That would be an ultimate on the fly test to me.
Picking up a bundle of loose pipes actually seems like a great benchmark for humanoid robots. Especially if they're not in a perfect pile. A full test could be something like grabbing all the pipes, from the floor, and putting them into a truck bed, in some (hopefully) sane fashion
I have my personal multimodal benchmark for physical robots.
You put a keyring with bunch of different keys in front of a robot and then instruct it pick it up and open a lock while you are describing which key is the correct one. Something like "Use the key with black plastic head and you need to put it in teeths facing down"
I have low hopes of this being possibe in the next 20 years. I hope I am still alive to witness if it ever happens.
I suppose the next big milestone is Wozniak's Coffee Test: A robot is to enter a random home and figure out how to make coffee with whatever they have.
Is that a real 5-10 years, a research 5-10 years[0] or 5-10 years of "FSD in the next 6 months"?
The demo space is so sterile and empty I think we're still a loong ways off from the Coffee test happening. One big thing I see is they don't have to rearrange other items they have nice open bins/drawers/shelves/etc to drop the items into. That kind of multistep planning has been a thorn in independent robotics for decades.
Natural, cluttered environments are a lot tougher to deal with. This near future-y minimalist environment has the dual benefits of looking stylish and being much closer to whatever they were able to simulate at scale for training the models.
At this point, this is enough autonomy to have a set of these guys man a howitzer (read as old stockpiles of weapons we already have). Kind of a scary thought. On one hand, I think the idea of moving real people out of danger in war is a good idea, and as an American i'd want Americans to have an edge... and we can't guarantee our enemies won't take it if we skip it, on the other hand I have a visceral reaction to machines killing people.
I think we're at an inflection point now where AI and robotics can be used in warfare, and we need to start having that conversation.
I don't understand we already saw exactly what happens with the emergence of drones and Israel is already using AI to select bombing targets and semi-autonomous turrets. What conversation? What kind of society do you think we are living in?
We had sufficient AI to make death machines for decades. You don't need fancy LLMs to get a pretty good success rate for targeting.
I have said for years that the only thing keeping us from "stabby the robot" is solving the power problem. If you can keep a drone going for a week, you have a killing machine. Use blades to avoid running out of ammo. Use IR detection to find the jugular. Stab, stab and move on. I'm guessing "traditional" vision algorithms are also sufficient to, say, identify an ethnicity and conduct ethnic cleansing. We are "solving the power problem" away from a new class of WMDs that are accessible to smaller states/groups/individuals.
> We had sufficient AI to make death machines for decades
And we already reached the peek here. Small drones that are cheaply mass produced, fly on SIM cards alone and explode when they reached a target. That's all there is to it. You don't need a gun mounted on a spot or a humanoid robot carrying a gun. Exploding swarms are enough.
So you're concerned about remote operated howitzers? Autoloaders and remote control land vehicles have existed for 40 or so years by now. If we wanted remote controlled howitzers we could have fielded them already.
Panzerhaubitze 2000 already has an autoloader and the entire point of self propelled artillery is that it moves after shooting to avoid counter artillery fire.
We already have an ongoing major war (in Ukraine) where both sides are using autonomous AI-driven drones that kill people, at scale, with escalating tit-for-tat advances in lethality. This conversation is being had alright.
I'm always wondering at the safety measures on these things. How much force is in those motors?
This is basically safety-critical stuff but with LLMs. Hallucinating wrong answers in text is bad, hallucinating that your chest is a drawer to pull open is very bad.
Battle fields are violent and exhausting, people get the shakes, make mistakes, hurt themselves and each other all the time. Munitions are generally designed to not explode from a bump in the road or getting dropped or squeezed, and targeting systems commonly support automatic tracking or similar, for this specific reason.
In terms of low-level safety, they can probably back out forces on the robot from current or torque measurement and detect collisions. The challenge comes with faster motions carrying lots of inertia and behavioral safety (e.g. don't pour oil on the stove)
That is how I would design it. It is common in safety critical PLC systems to have 1 or more separate safety PLCs that try to prevent bad things from happening.
As a sibling comment implies though, there's also danger from it being stupid while unsupervised. For example, I'd be very nervous having it do something autonomously in my kitchen for fear of it burning down my house by accident.
From a different robot (Boston Dynamics' new Atlas) - the system moves at a "reasonable" speed. But watch at 1m20s in this video[1]. You can see it bump and then move VERY quickly -- with speed that would certainly damage something, or hurt someone.
That's not enough. When a robot link is in motion and hits an object, the change in momentum creates an impulse over the duration of deceleration. The faster the robot moves, the faster it has to decelerate, the higher the instantaneous braking force at the impact point.
That's actually more of a solved problem. Robot arms that can track the force they're applying and where to avoid injuring humans have been kicking around for 10-15 years. It let them go out of the mega safety cells into the same space as people and even do things like letting the operator pose the robot to teach it positions instead of having to do it in a computer program or with a remote control.
The term I see a lot is co-robotics or corobots. At least that's what Kuka calls them.
That's fine for wheeled robots or robots bolted to the floor but for legged robots, especially bipeds, the hard question is how to prevent them from falling over on things. These don't look heavy enough to be too dangerous for a standing adult but you've still got pets/children to worry about.
177 comments
[ 3.5 ms ] story [ 317 ms ] threadWhat is the interface from the top level to the motors?
I feel it can not just be a neural network all the way down, right?
huh. An interesting approach. I wonder if something like this can be used for other things as well, like "computer use" with the same concept of a "large" model handling the goals, and a "small" model handling clicking and stuff, at much higher rates, useful for games and things like that.
200Hz is barely enough to control a motor, but it is good enough to send a reference signal to a motor controller. Usually what is done is that you have a neural network to learn complex high level behaviour and use that to produce a high level trajectory, then you have a whole body robot controller based on quadratic programming that does things like balancing, maintaining contacts when holding objects or pressing against things. This requires a model of the robot dynamics so that you know the relationship between torques and acceleration. Then after that you will need a motor controller that accepts reference acceleration/torque, velocity and position commands which then is turned into 10kHz to 100kHz pulse width modulated signals by the motor controller. The motor controller itself is driving MOSFETs so it can only turn them on or off, unless you are using expensive sinusoidal drivers.
I could also imagine a lot of safety around leaving things outside of the current task alone so you might have to bend over backwards to get new objects worked on.
These models are trained such that the given conditions (the visual input and the text prompt) will be continued with a desirable continuation (motor function over time).
The only dimension accuracy can apply to is desirability.
However, as it was trained using generic text data similarly to a normal LLM, it knows how an apple is supposed to look like.
Similar than a kid that never saw a banana, but his parent described it to him.
Vision+language multimodal models seem to solve some of the hard problems.
Do you really expect the oligarchs to put up with the environmental degradation of 8 billion humans when they can have a pristine planet to themselves with their whims served by the AI and these robots?
I fully anticipate that when these things mature enough we'll see an "accidental" pandemic sweep and kill off 90% of us. At least 90%.
Fortunately, robotic capability like that basically becomes the equivalent of Nuclear MAD.
Unfortunately, the virus approach probably looks fantastic to extremist bad actors with visions of an afterlife.
I'd rather have less waged labour and more time for chores with the family.
https://www.aliexpress.com/w/wholesale-clothes-folding-machi...
Why make such sinister-looking robots though...?
But it did seem like title of their mood board must have been "Black Mirror".
Very uncanny valley, the glossy facelessness. It somehow looks neither purely utilitarian/industrial nor 'friendly'. I could see it being based on the aesthetic of laptops and phones, i.e. consumer tech, but the effect is so different when transposed onto a very humanoid form.
When I hire a plumber or a mechanic or an electrician, I'm not just paying for muscle. Most of the value these professionals bring is experience and understanding. If a video-capable AI model is able to assume that experience, then either I can do the job myself or hire some 20 year old kid at roughly minimum wage. If capabilities like this come about, it will be very disruptive, for better and for worse.
If you can't get a tutorial on your exact case you learn about the problem domain and intuit from there. Usually it works out if you're careful, unlike software.
Case in point, I remember about ten years ago our washing machine started making noise from the drum bearing. Found a Youtube tutorial for bearing replacement on the exact same model, but 3 years older. Followed it just fine until it was time to split the drum. Then it turned out that in the newer units like mine, some rent-seeking MBA fuckers had decided more profits could be had if they plastic welded shut the entire drum assembly. Which was then a $300 replacement part for a $400 machine.
An AI doesn't help with this type of shit. It can't know the unknown.
Or you could wear it while you cook and it could give you nutrition information for whatever it is you cooked. Armed with that it could make recommendations about what nutrients you're likely deficient in based on your recent meals and suggest recipes to remedy the gap--recipes based on what it knows is already in the cupboard.
But none of the kitchen stuff we learned had anything to do with ensuring that this week's shopping list ensures that you'll get enough zinc next week, or the kind of prep that uses the other half of yesterday's cauliflower in tomorrow's dinner so that it doesn't go bad.
These aren't hard problems to solve if you've got time to plan, but they are hard to solve if you are currently at the grocery store and can't remember that you've got a half a cauliflower that needs an associated recipe.
Presumably, they won't as this is still a tech demo. One can take this simple demonstration and think about some future use cases that aren't too different. How far away is something that'll do the dishes, cook a meal, or fold the laundry, etc? That's a very different value prop, and one that might attract a few buyers.
In other words: I'm sorry, but that's how reality turned out. Robots are better at thinking, humans better at laboring. Why fight against nature?
(Just joking... I think.)
I'd totally use that to clean my garage so that later I can ask it where the heck I put the thing or ask it if I already have something before I buy one...
I think the key point why this "reverse cyborg" idea is not as dystopian as, say, being a worker drone in a large warehouse where the AI does not let you go to the toilet is that the AI is under your own control, so you decide on the high level goal "sort the stuff away", the AI does the intermediate planning and you do the execution.
We already have systems like that, every time you use you tell your navi where you want to go, it plans the route and gives you primitive commands like "on the next intersection, turn right", so why not have those for cooking, doing the laundry, etc.?
Heck, even a paper calendar is already kinda this, as in separating the planning phase from the execution phase.
For "stuff" I think a bigger draw is having it so it can let me know "hey you already have 3 of those spices at locations x, y, and z, so don't get another" or "hey you won't be able to fit that in your freezer"
This can be done of course, in your statement the phrase “just figure out” is doing a lot more heavy lifting than you allude to
Its similar to losing callouses on our hands if you don’t labor/go to the gym.
> Manna told employees what to do simply by talking to them. Employees each put on a headset when they punched in. Manna had a voice synthesizer, and with its synthesized voice Manna told everyone exactly what to do through their headsets. Constantly. Manna micro-managed minimum wage employees to create perfect performance.
[0] https://marshallbrain.com/manna1
EDIT: Let alone chop an onion. Let me tell you having a robot manipulate onions is the worst. Dealing with loose onion skins is very hard.
I now they claim there’s no special coding but did they practice this task? Special training?
Even if this video is totally legit I’m but burned out by all the hype videos in general.
The latest models seemed to be fluidly tied in with generating voice; even singing and laughing.
It seems like it would be possible to train a multimodal that can do that with low level actuator commands.
If that sounds like a cheat, neuroscientists tell us this is how the human brain works.
You put a keyring with bunch of different keys in front of a robot and then instruct it pick it up and open a lock while you are describing which key is the correct one. Something like "Use the key with black plastic head and you need to put it in teeths facing down"
I have low hopes of this being possibe in the next 20 years. I hope I am still alive to witness if it ever happens.
The demo space is so sterile and empty I think we're still a loong ways off from the Coffee test happening. One big thing I see is they don't have to rearrange other items they have nice open bins/drawers/shelves/etc to drop the items into. That kind of multistep planning has been a thorn in independent robotics for decades.
[0] https://xkcd.com/678/
Does anyone know if this trained model would work on a different robot at all, or would it need retraining?
I think we're at an inflection point now where AI and robotics can be used in warfare, and we need to start having that conversation.
I have said for years that the only thing keeping us from "stabby the robot" is solving the power problem. If you can keep a drone going for a week, you have a killing machine. Use blades to avoid running out of ammo. Use IR detection to find the jugular. Stab, stab and move on. I'm guessing "traditional" vision algorithms are also sufficient to, say, identify an ethnicity and conduct ethnic cleansing. We are "solving the power problem" away from a new class of WMDs that are accessible to smaller states/groups/individuals.
And we already reached the peek here. Small drones that are cheaply mass produced, fly on SIM cards alone and explode when they reached a target. That's all there is to it. You don't need a gun mounted on a spot or a humanoid robot carrying a gun. Exploding swarms are enough.
This is basically safety-critical stuff but with LLMs. Hallucinating wrong answers in text is bad, hallucinating that your chest is a drawer to pull open is very bad.
Battle fields are violent and exhausting, people get the shakes, make mistakes, hurt themselves and each other all the time. Munitions are generally designed to not explode from a bump in the road or getting dropped or squeezed, and targeting systems commonly support automatic tracking or similar, for this specific reason.
There cannot be a safety system of this type for a generalist platform like a humanoid robot. It's possibility space is just too high.
I think the safety governor in this case would have to be a neural network that is at least as complex as the robots network, if not more so.
Which begs the question: what system checks that one for safety?
it's easy to take your able body for granted, but reality comes to meet all of us eventually.
[1] https://www.youtube.com/watch?v=F_7IPm7f1vI
The term I see a lot is co-robotics or corobots. At least that's what Kuka calls them.
The article clearly spells out that it's end to end LLM. Text and video in, motor function out.
Technically, the text model probably has a few copies, but they are nothing more than Asimov's narrative. Laws don't (and can't) exist in a model