This is an interesting list, and I believe Tesla can develop these robots and create a market for them. On the other hand, I am slightly skeptical about how the robot will ultimately handle these tasks right now and on what schedule learning new tasks will occur.
Additionally, practically every one of these examples is one where the robot has to perform a surprisingly demanding task. Perhaps the most striking to me was the one you mentioned: pushing blanks to industrial robots. A routine and boring task, but is it really? Many tasks in factories are surprisingly complex (at least if the job description is expanded slightly) and varied, requiring a lot of tacit knowledge. My own experience, having seen a few factories, is that warehouse tasks, among others, are surprisingly demanding. For example, palletizing, stacking goods on a pallet, placing the pallet labelâand always exactly on the right palletâwrapping the pallets in film, strapping, and moving them from place to place is in some parts at quite a primitive level in companies. Furthermore, itâs often handled by a single warehouse worker, and if they arenât there, the job stops right there or is done as expensive overtime.
Regarding my faith in the future of Teslaâs robots, Iâm not for or againstâbased on everything Iâve seen, rather the opposite. I would very much like them to succeed, but outside of Teslaâs factories, there may be a surprisingly steep learning curve ahead.
The best use case for humanoid robots is as professional athletes. All sports, from football and ice hockey to basketball and baseball, can easily be taught to humanoid robots. In any sport, the mechanics are quite straightforward. Robots are superior in speed, strength, jumping, etc. Furthermore, robots can have 360-degree vision and process information continuously without their brains getting tired.
They can be recharged during intermissions or shifts, so the short operating timeâwhich is a drawback for humanoid robots in other fields of workâwould not be an issue.
There are dozens of sports clubs in the world with team values exceeding $1B, and thousands of smaller ones worth over $M. Additionally, every year players are bought/traded for at least hundreds of $M, so the turnover for robot manufacturing would be stable and very high.
I donât believe in Muskâs visions at all (even with the proverbial grain of salt). SpaceXâs mission to colonize Mars is silly, to say the least. Instead, hauling cargo into orbit is a trillion-dollar business, and in that, SpaceX is number one. Humanoid robots are also some kind of romantic vision. Robots exist and will come in various forms, but in the near future, they wonât have legs on a large scale. A human-shaped robot with human faces loses quite a bit of productivity compared to an optimized solution.
Quite the opposite. Sports clubs wouldnât be worth billions if humanoid robots replaced athletes. This argument completely fails to understand where the cash flows of sports clubs come from and what makes people follow sports. It is precisely those individuals who are exceptionally good at their sportsânot robots capable of a theoretically perfect performance.
A humanoid robot is not intended to be the most efficient solution for restocking, but rather to create a sense of security for people shopping in the store while the restocking takes place. If some âtentacle boxâ is moving around in the aisle, the customerâs mind wonât be at ease because the movements are efficient, but unfamiliar.
Of course, a humanoid robot could twist its limbs in any which way and move unexpectedly, but these should be avoided so that people do not become anxious.
Who would watch sports between robots? Computers have been better than humans at chess for 30 years, but no one is interested in chess between computers. And a floor jack beats the worldâs best powerlifters, but who would be interested in watching which robot lifts the most iron?
And as for wheels, it is by far the most efficient way to move a humanoid robot. A robot can be capable of walking stairs and still move on wheels. It makes no sense for robots to have two large feet when they could have wheels in their place that can also be used for walking. Or wheels on the soles of the feet, and the robotâs speed and energy efficiency will multiply.
Irony is certainly a difficult art on discussion forums. I donât know what the original author meant with their comment about humanoid robots and professional athletes, but I at least read it as clear-as-day irony.
Iâd actually be interested in something like Robot Wars (which used to air on Sub) featuring modern AI robots, maybe pitting them against each other in an octagon. Too bad Musk burned his bridges with the White House, otherwise it could have been organized there.
You could imagine itâs more sensible and cost-effective to churn out a single (humanoid-style) robot thatâs less productive but versatile, rather than building a custom one for every single purpose. These probably arenât even designed to replace or compete with things like welding arms on a car assembly line.
We will certainly see various robot events for entertainment purposes. For instance, in racket sports, you could have a great intermission show by having a robot play tennis against someone like Alcaraz.
One podcast interviewed Musk for three hours about the development of Optimus and many other topics. According to Musk, there are three major challenges for Optimus:
Human-like hand. The entire motor system had to be built in-house as there are no off-the-shelf components. Optimus 3 has 22 degrees of freedom, whereas a human hand has 27 DOF.
Real-world intelligence. Even though FSD has provided a good foundation, understanding the physical environment still requires new machine learning (slipperiness, the weight of objects, predicting physical movements).
Scaling. Since many components (e.g., related to the hand) are proprietary, scaling is much more laborious than buying parts from the market.
To address the challenges of real-world understanding, Musk describes âOptimus Academy,â where thousands of robots learn a specific task, and successes are recorded in a shared neural network. Additionally, there is a simulation environment similar to FSD, where virtual robots train in a simulated environment that includes various real-world surprises, such as oil on the floor.
Just for comparison, Amazon has a million robots in its warehouse operations, and they also have the humanoid robot Digit, which doesnât have human-like fingers but is capable of stacking boxes. Itâs unlikely this would be trialed if other industrial robots were capable of doing the job.
I asked about this earlier, but I donât remember anyone answering. The usefulness of industrial robots operating among people is limited by restrictions on their force and movement speed. These are set for safety reasons, so the robot doesnât knock down a person in its path.
Are there any standards for the force usage and movement speed of robots sold for home use, or can a robot, for example, twist open a tin can as hard as necessary? Or swing a mop at a speed that causes injuries to a person or pet in its way?
I donât see this kind of regulation as an obstacle to robotics, but rather as a safety net for the manufacturer that can be cited when accidents inevitably occur. Unless there are publicly referable limits for the movement speed and force of home robots, we can expect some interesting lawsuits.
Robot soccer as a programming competition with identical robots is absolutely hilarious to watch.
I do understand your point that these are technology demos and practice projects, and I donât see a very large market for sports robots as a commercial product myself either.
Is the bolded part a fact or an opinion? Iâm asking because I donât know. In any case, human sports will likely continue to be followed, but competition between robots could certainly emerge alongside it in some form. Even F1 racing is ultimately a contest between car manufacturers.
Of course itâs not a coincidence, as this is the only âform factorâ that investors are willing to pour money into without any kind of market research or profitability calculations. Because as stated, the goal is for a humanoid robot to be able to perform a wide variety of tasks just like a human, which presumably means there are enough suitable use cases without having to think about them beforehand. A high risk, high reward game. In the design phase of a typical industrial robot, on the other hand, it is known what the robot needs to be able to do and what customers are willing to pay for it â low risk, low reward.
There are all sorts of other forms in development, and I was mainly referring to this wheels vs. legs debate. At this stage, legs seem to be what many big players are focusing on; perhaps that will change in the future. I wouldnât include traditional industrial robots in this discussion at all, as they are products for a completely different segment and for completely different needs.
Thatâs how developing anything new tends to be, and nothing guarantees that Tesla will take the jackpot here, even though they seem to be in a good position. AI will play a decisive role in everything from now on, so one needs to succeed with it, which seems to mean the ability to invest sufficiently and move quickly to scale capacity and/or come up with new innovations on the software side.
Yeah, and legs are chosen over wheels to keep options open. If it were being made for a strictly defined need, the best form for that specific need would be chosen. But since they donât want to exclude any human labor, a human form is used. Thatâs what I was getting at with the industrial robot comparison â there, they know the purpose the robot is for and design the form accordingly.
In the development of humanoid robots, the goal is to build a human-like robot. The objective is not really about a robot intended for practical utility, but rather the human desire to build a robot that resembles a human as closely as possible. Thus, the goal is more of an entertainment pursuit and a competition of technical capability.
A human-like robot has always been a kind of fantasy for people and has been prominently featured in films based on sci-fi fantasies, among others. In its own way, the humanoid robot is a measure of technical capability, and building one has become a competitive goal in itself.
Perhaps in that race, there is even a sort of âbiblicalâ goal of creation and competition (God created man in his own image).
A humanoid robot itself is not the most efficient model for a robotic servant, but through the competition to build one, technical development in robotics advances rapidly, and these achievements are put to practical use in industrial robotics.
The wheels vs. legs debate is somewhat silly, as everyone realizes that the wheel is by far the most efficient way to move (on land), especially from an energy economy standpoint. Even a humanoid robot needs energy to operate, and therefore energy consumption must be minimized. If a humanoid robot needs a charging cable up its rear after walking every flight of stairs, its assistance with household chores will be quite slow-paced.
A bicycle is certainly not the best or most efficient locomotion technology on land. If you try using a bicycle in a forest or even in open terrain, even a grandmother will walk past you faster. A bicycle is good on flat and predictable surfaces.
The best, however, is a combination of legs and wheels. You use wheel drive as long as itâs flat and clear obstacles by leaping and jumpingâthat is, legwork combined with wheel drive. It is a very fast and fluid way to travel. Downright frighteningly smooth, based on the demos seen in videos.
Too bad Musk burned his bridges with the White House, otherwise it could have been organized there.
