Robots are hard says Elon Musk

90 views
Skip to first unread message

Thomas Messerschmidt

unread,
Jul 24, 2026, 2:09:22 AMJul 24
to hbrob...@googlegroups.com

Chris Albertson

unread,
Jul 24, 2026, 9:45:56 PMJul 24
to hbrob...@googlegroups.com
I saw that video of the popcorn-dispensing robot.    It’s sad.   The robot is remotely controlled by a human with a notebook computer just a few feet away.   How many billion dollars did Tesla spend on that?


What I hope for is that we only need one smart guy to think of some revolutionary idea.  He need not work for a billion-dollar company.  It could be anyone.   

Remember the paper “All you need is attention?”    That one guy, a student at the time, revolutionized the entire AI industry and enabled the current era of LLMs.

Like Newton, who had just one idea: “What if gravity’s pull extended all the way up into space, to the Sun and the Moon?” Everyone at the time assumed gravity stopped working maybe only a few miles above the Earth’s surface, and it was “obvious" because the Moon did not fall to the ground. But it only takes one person to ask differently.

Tesla and all the rest of us will be stuck with remote-controlled popcorn machines for a long time. Maybe Yann LeCun's concept of world models will pay off. But I’ve also thought we have the answer in plain sight: brain waves. You have consciousness if you have brain waves, and when they stop, they call you “brain dead.” So far, all AI must be brain dead. That should be a clue.

Musk is 100% correct:  Building good robots with 2026 technology is almost a lost cause.   We can build good-enough walking machines, but they are mindless zombies.




--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/hbrobotics/E511340A-1117-4DE9-9F2B-AF6EB2D51999%40gmail.com.

Wayne Gramlich

unread,
Jul 24, 2026, 10:10:48 PMJul 24
to hbrob...@googlegroups.com, Wayne Gramlich
Chris:

The paper "Attention Is All You Need" is one of the most cited papers on the planet. I'm curious why you think Aidan Gomez, who was a student at Univ. of Toronto at the time, is the primary person responsible for the paper?  I'm just curious.

Regards,

-Wayne

David Murphy

unread,
Jul 24, 2026, 10:44:45 PMJul 24
to hbrob...@googlegroups.com
All authors on that paper are credited with equal contribution…from the paper itself…

Equal contribution. Listing order is random. Jakob proposed replacing RNNs with self-attention and started

the effort to evaluate this idea. Ashish, with Illia, designed and implemented the first Transformer models and

has been crucially involved in every aspect of this work. Noam proposed scaled dot-product attention, multi-head

attention and the parameter-free position representation and became the other person involved in nearly every

detail. Niki designed, implemented, tuned and evaluated countless model variants in our original codebase and

tensor2tensor. Llion also experimented with novel model variants, was responsible for our initial codebase, and

efficient inference and visualizations. Lukasz and Aidan spent countless long days designing various parts of and

implementing tensor2tensor, replacing our earlier codebase, greatly improving results and massively accelerating

our research.



--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.

Chris Albertson

unread,
Jul 24, 2026, 11:18:00 PMJul 24
to hbrob...@googlegroups.com, Wayne Gramlich
I think because I confused the date with something else that came from Stanford.  I wasn’t a typo kind of thing, I’ve remembered this wrong for quite a while.   Yes, the paper came from Google.  I’d remembered wrongly long enough that I was sure of it so I did not bother to check.   It is like confusing two big murals: “The Last Judgment” and “The Last Supper”.    Da Vinci did one, Michelangelo, the other.   

But still, I wanted to show how the field of AI is so new that all the best discoveries are still likely in the future, and it is wide open still to the point where anyone could still discover something.

BTW, I think (guess) attention works because the brain is not a state machine.  We store the last few seconds as an echo and then have “detectors” that somehow fire when they see the right combination of echoes.  Attention kind of dumps the whole input in at once (yes it does encode word order but that is just one of many things). That was a big deal.    

I think current robots are so bad because some large piece of the puzzle is missing, not because engineers need to work harder and longer.   So even with a billion more dollars will not build a robot that gives you popcorn because it feels that you might like it.  It serves because the man across the room clicked a button.

Wayne Gramlich

unread,
Jul 24, 2026, 11:28:31 PMJul 24
to Chris Albertson, hbrob...@googlegroups.com, Wayne Gramlich
Chris:

I was just wondering if you had some additional information about that
paper that literally transformed (sorry, I couldn't resist ;-) AI.

I will point out that the human brain appears to use somewhere in the range
of 20-2 watts, but require 2+ decades to train.  So, there is clearly much more
opportunity for improvement to AI algorithms.

Regards,

-Wayne

Dave Everett

unread,
Jul 25, 2026, 12:36:41 AMJul 25
to hbrob...@googlegroups.com
The robotics merry-go-round comes round again, just like it does every decade or so.

Musk admitting robotics is hard might be more than just an admission. It's clear Tesla are behind others, that is not surprising as they were a late starter, but mentioning teleoperation, when Tesla has done the same might be to poison the well. If doubt can be cast (and it can) on other companies, the public might hold out for Tesla's offering. Given Tesla's history, they are still a long way off. I was recently asked if I would assist with a home robot projects, one of the other people asked, suggested it would be futile, not because the problem was difficult, but because Tesla would surely be releasing their robot soon and it would dominate the market.

Every time the merry-go-round comes around, companies are trying to be the first to market, and that always results in a poor product. The strength is with us, the hobbyist, not beholden to shareholders, perhaps wives though. We can tinker away and sometimes we make breakthroughs as Chris says. 

Dave

Chris Albertson

unread,
Jul 25, 2026, 12:46:24 PMJul 25
to hbrob...@googlegroups.com


> On Jul 24, 2026, at 9:36 PM, Dave Everett <daveev...@gmail.com> wrote:


> The strength is with us, the hobbyist, not beholden to shareholders, perhaps wives though. We can tinker away and sometimes we make breakthroughs as Chris says.




While I hope some amateur might have a groundbreaking idea. I’m doubtful. The current state of the art is now at a point where one needs a considerable education just to understand the problems.

Hobbyists like to tinker with screwdrivers and mechanical parts and seem to actively avoid hard theoretical issues or collaboration. So I have my doubts. But on the other hand, it only takes one counterexample to prove me wrong. It would be great if I were wrong.

So here is a question? In theory, what would it take to build a robot that cleans a bathroom because it looks inside and thinks to itself, “this is disgusting, I need to do something about this,” rather than cleans the bathroom because someone clicked the “clean bathroom” button? In other words, a human-equivalent AI.

OK this is a hard problem with no certain payoff. You will likely fail. So it would seem perfect. for “the hobbyist, not beholden to shareholders”. But that is not what we see most work on. Hobbyists work on project using decades old technology and on projects they know will work.



Wayne Gramlich

unread,
Jul 25, 2026, 1:14:11 PMJul 25
to hbrob...@googlegroups.com, Wayne Gramlich
All:

I'm a tad more optimistic about hobbyist robotics...

I will point out that the Aloha arms (both the leaders and followers) were built out of Dynamixel servos
that were squarely targeted towards the Robotic hobbyist/educational markets.

If I were going to do a hobbyist humanoid robot, I would do the upper torso using servo technology and punt on servo based legs.  Instead a perfectly serviceable holonomic base can be assembled using off-the shelf holonomic wheels available from the FIRST robotics crowd.

Also, there is plenty of room for experimenting with improving touch sensors for hands.

Regards,

-Wayne

James H Phelan

unread,
Jul 25, 2026, 2:25:35 PMJul 25
to hbrob...@googlegroups.com
"So here is a question?   In theory, what would it take to build a robot that cleans a bathroom because it looks inside and thinks to itself, “this is disgusting, I need to do something about this,” rather than cleans the bathroom because someone clicked the “clean bathroom” button?    In other words, a human-equivalent AI."

That, in theory, is not that difficult. It's all in the Behavior Tree:
Routine home maintenance patrol. Go to bathroom. Inspect toilet. Does camera image resemble "clean toilet" or "dirty toilet"? If "dirty toilet" begin "clean toilet" routine. Re-inspect toilet. If clean apply "toilet duck" and move on, if not alert user to failure of "clean toilet" routine, place in debug ToDo list, and move on. Note - it may well be much less work to clean the toilet yourself than to debug why the robot failed to clean it. This will ever more likely to be the case as AI progresses: The thing messed up. We don't know why. It's not worth fixing at the moment. Deal with it. Paging Dr Calvin. Paging Dr Susan Calvin....

James H Phelan
"Nihil est sine ratione cur potius sit quam non sit"
Leibniz

Thomas Messerschmidt

unread,
Jul 25, 2026, 2:31:39 PMJul 25
to hbrob...@googlegroups.com
It would take 50 years of engineering. 



Thomas Messerschmidt

-  

Need something prototyped, built or coded? I’ve been building prototypes for companies for 15 years. I am now incorporating generative AI into products.

Contact me directly or through LinkedIn:   




On Jul 25, 2026, at 11:25 AM, 'James H Phelan' via HomeBrew Robotics Club <hbrob...@googlegroups.com> wrote:


"So here is a question?   In theory, what would it take to build a robot that cleans a bathroom because it looks inside and thinks to itself, “this is disgusting, I need to do something about this,” rather than cleans the bathroom because someone clicked the “clean bathroom” button?    In other words, a human-equivalent AI."

That, in theory, is not that difficult. It's all in the Behavior Tree:
Routine home maintenance patrol. Go to bathroom. Inspect toilet. Does camera image resemble "clean toilet" or "dirty toilet"? If "dirty toilet" begin "clean toilet" routine. Re-inspect toilet. If clean apply "toilet duck" and move on, if not alert user to failure of "clean toilet" routine, place in debug ToDo list, and move on. Note - it may well be much less work to clean the toilet yourself than to debug why the robot failed to clean it. This will ever more likely to be the case as AI progresses: The thing messed up. We don't know why. It's not worth fixing at the moment. Deal with it. Paging Dr Calvin. Paging Dr Susan Calvin....

James H Phelan
"Nihil est sine ratione cur potius sit quam non sit"
Leibniz
On 7/25/2026 11:46 AM, Chris Albertson wrote:
So here is a question?   In theory, what would it take to build a robot that cleans a bathroom because it looks inside and thinks to itself, “this is disgusting, I need to do something about this,” rather than cleans the bathroom because someone clicked the “clean bathroom” button?    In other words, a human-equivalent AI.

--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.

James H Phelan

unread,
Jul 25, 2026, 2:41:52 PMJul 25
to hbrob...@googlegroups.com

I'm astonished that, because of shared information, solutions that are "10 years into the future" with linear arithmetic projections are, with truly "exponential" synergy, announced the following month!

James H Phelan
"Nihil est sine ratione cur potius sit quam non sit"
Leibniz

Chris Albertson

unread,
Jul 25, 2026, 3:23:12 PMJul 25
to hbrob...@googlegroups.com
On Jul 25, 2026, at 11:25 AM, 'James H Phelan' via HomeBrew Robotics Club <hbrob...@googlegroups.com> wrote:

That, in theory, is not that difficult. It's all in the Behavior Tree:
Routine home maintenance patrol. Go to bathroom. Inspect toilet. Does camera image resemble "clean toilet" or "dirty toilet"? If "dirty toilet" begin "clean toilet" routine. Re-inspect toilet. If clean apply "toilet duck" and move on, if not alert user to failure of "clean toilet" routine, place in debug ToDo list, and move on. Note - it may well be much less work to clean the toilet yourself than to debug why the robot failed to clean it. This will ever more likely to be the case as AI progresses: The thing messed up. We don't know why. It's not worth fixing at the moment. Deal with it. Paging Dr Calvin. Paging Dr Susan Calvin....

The behavior tree only automates the button pressing by comparing the amount of dirt to some threshold.  The robot never actually feels disgusted at an emotional level and is not self-motivated to “do something”.   It is at best a mindless zombie.

The behavior tree will never invent a novel solution to a problem.

That is the fundamental difference between a toaster oven and a human.   The human can be the “first cause,” and he knows that it “feels like something” to be a human.   Robots running behavior trees don’t have that.

But your solution only works for bathrooms.  Even if you wrote a special “if-this-then-then"logic for a million different things, you still do not solve the general case.   They tried this in the 1980s and worked for 30 years before giving up.   Remember the “Cyc” project?   They tried to hand-code every known fact in the universe and proved it was impossible.   I was all-in on Cyc in the 80s. My graduate adviser worked on it.  

I made the problem impossible hard on purpose.  It has been unsolved for generations.

What we need is Susan Calvin.  She invented a completely new technology that had general intelligence and could act on its own.  Asimov knew about computers and he purposely did not base his robots on advanced computers.   Positronic Brains were not computers. Maybe some kind of quantum mechanical thing, he never said.

Chris Albertson

unread,
Jul 25, 2026, 3:50:15 PMJul 25
to hbrob...@googlegroups.com

On Jul 25, 2026, at 11:41 AM, 'James H Phelan' via HomeBrew Robotics Club <hbrob...@googlegroups.com> wrote:

I'm astonished that, because of shared information, solutions that are "10 years into the future" with linear arithmetic projections are, with truly "exponential" synergy, announced the following month!

James H Phelan
"Nihil est sine ratione cur potius sit quam non sit"
Leibniz
On 7/25/2026 1:31 PM, Thomas Messerschmidt wrote:
It would take 50 years of engineering. 


I’d argue differently.     I ask how long it would take 19th-century engineers to make a train locomotive that could fly across the ocean.    The answer is “never”.  You could never incrementally improve a steam locomotive bit by bit to make it fly.   We had to wait for a new and different technology.

Same with our current AI. Progress comes in jumps when we abandon the old and invent something new. Improvements are nice but by definition not revolutionary.

Maybe in 50 years someone will invent something new, but that is not 50 years of engineering.  That is “in 50 years we toss all this LLM stuff in the trash and go with something new."


In other news, I think I have identified a way to connect something like a joystick to an RL-trained robot.   I was very stumped on how a trained neural network could connect to a hand-coded world like ROS2.   Not “new tech” just “new to me tech”.    Training goes slow.    I do a run overnight and see that I made a dumb software error, fix it then a day or so later find the next bug.

James H Phelan

unread,
Jul 25, 2026, 3:54:17 PMJul 25
to hbrob...@googlegroups.com

The problem as presented was "recognizing a toilet that needs cleaning",

not "have the human emotion of disgust", not "be a human",

nor "recognize and find a solution for every possible problem you could ever come across whether or not it's in your job description". In a human that directive would create an obsessive compulsive psychosis of suicidal proportion.

Don't be so sure "the behavior tree will never invent a novel solution to a problem". Reinforcement learning AIs have developed game strategies that no human had tried before.  Given sufficient agency, the robot could query the accumulated experience of [insert large number here] of its fellow robots to see if another had a solution. Or query the [another large number] YouTube help videos.

If you want "artificial emotion" you can create a "disgust score" if certain conditions are met.

One needn't BE a human to accomplish human goals any more than an airplane has to flap its wings.

James H Phelan
"Nihil est sine ratione cur potius sit quam non sit"
Leibniz
--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.

Thomas Messerschmidt

unread,
Jul 25, 2026, 4:00:59 PMJul 25
to hbrob...@googlegroups.com
I hope I’m wrong. 



Thomas Messerschmidt

-  

Need something prototyped, built or coded? I’ve been building prototypes for companies for 15 years. I am now incorporating generative AI into products.

Contact me directly or through LinkedIn:   




On Jul 25, 2026, at 11:41 AM, 'James H Phelan' via HomeBrew Robotics Club <hbrob...@googlegroups.com> wrote:



A J

unread,
Jul 25, 2026, 5:47:53 PMJul 25
to HomeBrew Robotics Club

I think that a few years ago the OpenAI CEO was saying in an interview that the rough plan was to have some 3D awareness in release 6 and much more by release 7 (not an accurate recollection). But today the LLM is not trained to understand physics in a video of a rock and feather falling through the air. If we could get a low-millisecond interpretation of stereo vision it might be easier. But I think it is very hard to train large-scale models to do this; it might take some tricks and a little luck.

Thomas Messerschmidt

unread,
Jul 25, 2026, 6:03:08 PMJul 25
to hbrob...@googlegroups.com
I think you have the right idea. If we could train physical AI with video that could be the leg up needed. 

LLMs are the low hanging fruit of AI. The vast amount of data required was there for the taking. (Stealing?) No such data exists for physical AI. As such, training AI for robots is a slow and expensive (human) labor intensive process. 



Thomas Messerschmidt

-  

Need something prototyped, built or coded? I’ve been building prototypes for companies for 15 years. I am now incorporating generative AI into products.

Contact me directly or through LinkedIn:   




On Jul 25, 2026, at 2:47 PM, A J <aj48...@gmail.com> wrote:



Dave Everett

unread,
Jul 25, 2026, 8:44:41 PMJul 25
to hbrob...@googlegroups.com
On Sun, 26 Jul 2026 at 02:46, Chris Albertson <alberts...@gmail.com> wrote:


So here is a question?   In theory, what would it take to build a robot that cleans a bathroom because it looks inside and thinks to itself, “this is disgusting, I need to do something about this,” rather than cleans the bathroom because someone clicked the “clean bathroom” button?    In other words, a human-equivalent AI.

Oddly, I was thinking about it only this morning.Mainly because my bathroom is disgusting.

I didn't think about it for long, but like any problem it is layers of problems. A friend of mine once said "you eat the elephant one bit at a time". This particular elephant will have to be handled with rubber gloves. However, the breakdown is the beginning of solving all the problems.

Joe Engleberger was working on this, you ,might recall video of his robot cleaning a toilet bowl. One of my concerns was hygiene, the robot's "gloves" are going to need further cleaning and replacing from time to time. Who does that? Can the robot handle this. Even our vacuum robots seem to need human intervention, even the ones that deposit their contents automatically require a human to empty that receptacle.

Starting with a toilet bowl, that problem is solvable, sure it will require a lot of work, especially if we are requiring the robot to check the quality of it's work, and I assume we are here.

Most human cleaners will clean regardless, and I don't think that is unreasonable to have the robot just clean anyway, as long as it can check it's work. It's not every day you experience explosive diarrhea, but when you do, you want the robot to do the job properly not just go through the "movements" so to speak.

In the short term, special attachments for a toilet brush etc is also reasonable. 

I think being able to tell the quality of the clean of a toilet bowl is also doable. This might require a camera that can zoom in so the robot could check for discolourations to further attack.

Breaking it down, we probably have some of the problems solved already:
1. How to get to the bathroom
2. Identify items to clean, toilet, basin, shower etc
3. Identify cleaning tools
4. manipulate cleaning tools
5. assess job

Yes these are just high level, further breakdown is required.

It seems to me that multiple arms are required just to get started. Opening a bleach bottle can't be done with one hand. Many have safety caps that require you to squeeze on two points while turning.  I'm not going to suggest a human do this for the robot, because if I do, it's short journey to the human doing everything and we are back as square one. Perhaps an automated caddy is needed to augment the robot's capabilities. The caddy could rotate round the correct item, and could remove the cap, or even dose the correct amount to the robot. It's not unreasonable at this stage to distribute the robotic tools throughout a home. The image of a robot carrying in a mop and a bucket filled with cleaning bottles is unrealistic right now, but ultimately where it should be. That is, as little modification to the home as possible, preferably none.

A robot could use barcodes to determine what cleaning solution it is manipulating. It could use the weight to tell if it needs to be reordered. It could measure out amounts for use and it could accidentally bleach your Ken Done bath mat.

OK this is a hard problem with no certain payoff.  You will likely fail.   So it would seem perfect. for “the hobbyist, not beholden to shareholders”.   But that is not what we see most work on.  Hobbyists work on project using decades old technology and on projects they know will work.       

I see people working on all kinds of things, not just stuff they know will work. I've seen a few attempts at cooking robots. They have been disastrous but they are having a go. Hobbyists work on farming robots, weeding robots, picking up socks, the list goes on.

Dave

Alan Downing

unread,
Jul 26, 2026, 1:07:58 AMJul 26
to hbrob...@googlegroups.com
Interesting stats: over 140 Chinese companies are building humanoid robots.  Since late 2024, Chinese cities have put together investment funds for humanoid robots worth over $26 billion (per Morgan Stanley.)

BYD is about to show off their new humanoid robot next month:

--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.

Carl Sutter

unread,
Jul 26, 2026, 3:13:04 PMJul 26
to HomeBrew Robotics Club
It seems like one advantage that animals have is a lot of touch sensors.  Our robots have some in the joints, and sometimes fingers, but nothing like we do.  Using vision alone is not nearly enough for fine motion except for limited tasks.  There is a rare condition called Anaphia that is similar to this, and very difficult for those people to do some basic things.

Chris Albertson

unread,
Jul 26, 2026, 5:19:08 PMJul 26
to hbrob...@googlegroups.com
The classic task that proves what you say is threading a screw into a nut by hand.  It has to be done by feel.

But I think the pressure sensors can be the joints if they are backdrivable.  We think we feel pressure on the fingers, but it could be that some force has caused the finger to move.

Not just hands but feet too.  I think we use foot pressure (amount other things) for balance.

All this can be explored in simulation

On Jul 26, 2026, at 12:13 PM, Carl Sutter <cfsu...@gmail.com> wrote:

It seems like one advantage that animals have is a lot of touch sensors.  Our robots have some in the joints, and sometimes fingers, but nothing like we do.  Using vision alone is not nearly enough for fine motion except for limited tasks.  There is a rare condition called Anaphia that is similar to this, and very difficult for those people to do some basic things.

--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.
Message has been deleted

A J

unread,
Jul 26, 2026, 6:55:46 PMJul 26
to HomeBrew Robotics Club
There is some interesting working connecting the Brain waves and Vision or Muscle movement.

The OOPS factor has been recorded by EEG when the human sees a Robot make a mistake.

There are newer papers on this subject but this one is pretty clear. There are some industrial EEG

helmets now, maybe brain waves could work with attention.

Stephen Williams

unread,
Aug 2, 2026, 8:55:19 PM (11 days ago) Aug 2
to hbrob...@googlegroups.com, Carl Sutter

There are a number of approaches that can get a lot of touch sensing points into a robot.  Some of them are inexpensive and not too difficult to wire up.  One that I found at CES is a new Bosch barometric sensor that supports I3C.  At a tech conference in Santa Clara, talked to another company that has built a high-quality magnetic sensor fingertip that will be for sale soon: 4 sensors that measure pressure and shear with direction.

I have a goofy solution I'm going to be trying on my humanoid hand design that is extremely inexpensive, mostly low-tech, with about $10-20 of electronics for the whole hand.  With that, I hope to get up to 3 touch points for each fingertip, along with finger position, and maybe finger torque.


Stephen

On 7/26/26 12:13 PM, Carl Sutter wrote:
It seems like one advantage that animals have is a lot of touch sensors.  Our robots have some in the joints, and sometimes fingers, but nothing like we do.  Using vision alone is not nearly enough for fine motion except for limited tasks.  There is a rare condition called Anaphia that is similar to this, and very difficult for those people to do some basic things.
--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.

Stephen Williams

unread,
Aug 2, 2026, 9:10:55 PM (11 days ago) Aug 2
to hbrob...@googlegroups.com, Chris Albertson

Lol, I'm just reading this...

We need smartphone-like availability of humanoid robotic hardware.

And then we need to use those mainly as avatars: telerobotics with a modest amount of ML to make it easy, safe, and useful to remotely control.  And we'll capture everything as training data.

With enough avatar training data, we likely will rapidly converge on much better & broader robotic autonomy range.


Tesla & others are positioning humanoid robots as automobiles.  They are expected to be as expensive as a luxury vehicle.  They are expected to last as long as a luxury vehicle (implied).

Most humanoid robots won't be like automobiles, they will be like bicycles or riding lawnmowers: From cheap to modestly expensive, but nothing like an automobile or home.

The cameras, compute package, radio, etc. will be more, but that can be separate in various ways.  Maybe we'll snap in a high-end mobile phone for part of it, which is free in some sense.  Maybe we will have plug in modules that we periodically upgrade like mobile phones.  That package is largely separate from the mechanics, drive electronics, integrated senses (touch, proprioception), and battery.

We don't do this with automobiles for a few reasons, except for entertainment & maps where this has been ceded almost completely to mobile phones.  I think we will with robots.  For one thing, we'll want the cellular & satellite connectivity that mobile phones already have.  Plus camera, audio, and other integration.  People using mobile technology are going to win over anyone trying to echo that stack.  The compute module approach might work, but most examples of those that I am aware of do not anticipate plugging in a complete mobile phone sensor & antenna package.


Stephen

Reply all
Reply to author
Forward
0 new messages