Tesla Optimus, Figure 03 & Atlas | The Machines Coming for Every Job

Vídeo em Inglês, publicado pelo canal ASA Hacks.

About this content: The summary, key points and topics below are generated automatically from the video transcript. They reflect the content, opinions, claims and predictions presented in the video and do not represent a position, validation or fact-check by IA Dots.

Video summary

The video discusses the rise of humanoid robots moving from laboratories to factory floors, detailing their hardware components, software advancements through end-to-end neural networks, market players, current deployments, and the future economic and societal implications of general-purpose robotic labor.

Key points

• Humanoid robots are designed to operate in a world built for human bodies, utilizing two legs, arms, and hands to navigate existing infrastructure.
• Modern humanoids integrate advanced actuators, multi-degree-of-freedom hands with sensors, and proprioceptive senses to achieve human-like movement and balance.
• Software has shifted from traditional scripting to end-to-end neural networks and reinforcement learning trained largely in simulation, allowing fleets to share learned skills overnight.
• Key industry players like Tesla, Figure, Boston Dynamics, Agility, and various Chinese startups are deploying robots in factories and warehouses for commercial tasks.
• The future roadmap projects factory deployment scaling through the late 2020s, cost reductions potentially matching or beating minimum wage labor, and eventual entry into domestic home environments.

Topics

Humanoid robotsArtificial intelligenceAutomationFactory deploymentRobotics hardwareReinforcement learningEconomic impact

Video transcript

Transcript from available captions; it may contain errors.

Right now, in factories in California, South Carolina, and Ohio, there are workers who never sleep, never eat, and never ask for a raise. They walk on two legs, and they learn new jobs every single week. This is not science fiction anymore. The humanoid robot, a machine built in our image, has left the laboratory and walked onto the factory floor. Tesla, Figure, Boston Dynamics, and a wave of Chinese startups are racing to build millions of them. The question is no longer whether they will work alongside us. It is how fast, how cheap, and what happens when they come home. Here is the strange bet at the heart of this entire industry, that the best shape for a robot is a person. Every factory, warehouse, and home on Earth was designed for one specific machine, the human body. Our stairs are human-sized, our tools fit human hands, our doors, shelves, and car seats all assume two legs, two arms, and a head full of sensors at about 5 and 1/2 ft off the ground. A wheeled robot is cheaper and more stable. A robotic arm is stronger and more precise, but neither can climb your stairs or grab a mug designed for human fingers. The humanoid wager is simple. Instead of rebuilding the world for robots, build robots for the world. It is an enormously harder problem. Walking on two legs is physically a controlled fall, repeated hundreds of times a minute. For a robot, every step is a real-time calculation across dozens of motors and balance sensors. For decades, that calculation was too slow and too expensive. What changed is everything else. Batteries got denser, motors got stronger, and AI got good enough to learn walking the way a child does, by falling over millions of times in simulation. The dream is older than you think. Honda's ASIMO was walking stairs in the year 2000. What DARPA Robotics Challenge in 2015 showed how far there was to go. Robots toppling over on live television. A decade later, the falling is mostly over. So, let us open one up, figuratively, and see what a modern humanoid is made of. Start with the muscles, actuators. A humanoid like Tesla's Optimus carries around 40 of them. Motors at every major joint, each a marvel of packaging with torque densities rivaling human muscle. Boston Dynamics electric Atlas pushes further with joints that let it do backflips and spin its torso a full 360°. Movements no human body can match. Then the hands, the hardest part. A human hand has 27 degrees of freedom. Optimus's latest hands have 22 with fingertip sensors that feel slip, texture, and pressure. Figure's hands add cameras in the palms, so the robot can watch its own grip. This is what separates a robot that crushes an egg from one that cracks it into a bowl. The senses come next. Depth cameras and lidar build a 3D map of the world many times per second. Inertial sensors track tilt a thousand times a second. Force sensors in the feet feel the ground. The result is proprioception, the robot's sense of where its own body is. The same sense that lets you touch your nose with your eyes closed. And the skeleton matters. Every kilogram of a robot's own weight is a kilogram its motors must carry. So, designers use hollow structures, carbon composites, 3D printed titanium. Tesla's Optimus weighs just 57 kg, lighter than most humans it works beside. Power it all with a battery, 2 to 3 kWh, good for 4 to 16 hours, and you have a machine that walks, lifts 20 kg, and manipulates tools all day on one charge. Balance deserves its own mention. Humans balance using the vestibular system in the inner ear plus constant tiny corrections. Robots replicate this with model predictive control, simulating hundreds of possible next steps many times per second and choosing the one that keeps the center of mass over the feet. When pushed, a good humanoid does what you do. It takes a step. But a body is just expensive sculpture without the software. And this is where the real revolution is. The old way was scripting everything. Move joint A to angle B. It worked in cages doing one task forever. It fails the moment the world changes. The new way is end-to-end neural networks. Engineers show the robot thousands of examples, humans guiding it through tasks, and the network learns the mapping from what it sees to what its motor should do. Figure's Helix system takes camera input plus a spoken instruction like fold the laundry and outputs motor commands directly. No programming in between. Training happens mostly in simulation. A robot can attempt a task millions of times virtually in the time of one real attempt falling failing discovering what works. This is reinforcement learning, and it taught robots a walking grace that took decades of hand tuning to achieve before. Then comes the fleet effect. When one robot learns a new skill, every robot in the fleet downloads it overnight. A thousand robots learn in parallel and share everything. No human workforce in history could do that. The honest caveat, autonomy is still the hard part. Most dazzling demos involve some teleoperation. The unsolved problem is reliability. A robot that works 99% of the time still fails hourly on a factory line. Getting from demo to dependable is the entire game. Data is the fuel. Companies now employ fleets of human operators wearing VR headsets remotely guiding robots through thousands of tasks to generate training examples. Every hour of teleoperation becomes thousands of hours of learned skill. So, who is actually winning? Meet the contenders. Tesla's Optimus is the scale play. 173 cm, 57 kg, lifts 20. Deliberately unremarkable hardware because Tesla's bet is manufacturing. It retooled its Fremont factory to build robots targeting a million units a year at 20 to 30,000 dollars each. Over a thousand already work inside Tesla's plants. Commercial sales are targeted for late 2027. If anyone can make humanoids cheap, it is the company that already builds millions of machines a year. Figure AI is the deployment play. Its Figure 03 works today at BMW's Spartanburg plant handling sheet metal at better than 99% accuracy. Over 90,000 parts and counting. Its Bot Queue factory is rated for 12,000 robots a year and home trials are on the road map. Where Tesla promises scale tomorrow, Figure is clocking real hours today. Boston Dynamics electric Atlas is the capability play. 56 degrees of freedom, the industry's most, with joints rotating beyond human limits. The athlete, backflips, parkour, heavy lifting. Hyundai is preparing it for its factories. It will not be the cheapest. It is meant to be the most capable. Agility's Digit is the quiet workhorse. No head, no fingered hands, just legs and grippers optimized for moving totes in warehouses. 65,000 hours logged across nine facilities. Sometimes the winning move is not to build a human. It is to build half of one. And then there is China. Unitree's G1 costs around $16,000 and shipped 5,500 units last year. Dozens of Chinese firms are racing down the cost curve. This is becoming nation versus nation. Do not forget the specialists, either. Apptronik's Apollo targets elder care and health care. Neura's cognitive robots bring artificial skin that senses approach before contact. The field is far wider than the famous names. Strip away the demos. What are they actually doing right now? At BMW Spartanburg, figure robots place sheet metal parts into fixtures with 5-mm precision at speeds matching human workers. Inside Tesla's Fremont factory, over 1,000 Optimus units sort battery cells and carry parts, sharing improvements across the fleet overnight. In warehouses, Digit robots walk miles of aisles moving totes, the repetitive, injury-prone work with the highest turnover in the economy. The pattern is consistent. The first jobs are the ones humans are already abandoning. So, what happens next? The road maps agree. 2026 to 2027, the factory years. Low-volume production, enterprise customers, narrow supervised tasks. 2028 to 2030, the cost collapse. Every doubling of production cuts costs 15% to 20%. The same curve that made solar panels and batteries ordinary. At $20,000, a robot costs less than a year of minimum wage labor. The economics flip and adoption goes vertical. 2030 and beyond, the home. The final boss. A factory is structured. A home is chaos. Toys on the floor, pets underfoot. Every kitchen different. The dexterity and judgment for domestic life is an order of magnitude harder than the factory line. But doubt about whether it happens is gone from every road map. The humanoid workforce is a when, not an if. Consider what $20,000 labor means. Not just cheaper factories, but care for aging populations, disaster response in places too dangerous for humans, and construction in remote regions. Japan, with its shrinking workforce, is already planning around it. The economic implications reach far beyond the factory gate. An honest documentary asks the hard questions. Jobs first, the optimistic case. Robots take unwanted jobs, fill labor shortages, and new work appears maintaining and supervising them. The pattern of every automation wave. The pessimistic case. This time the machine is general purpose. A robotic arm replaced one task. A humanoid can be retrained for any task. Economists are genuinely divided. Safety second. A 60-kg machine among people must fail safe every time. Force-limited joints and collision detection help. Regulation has not caught up yet. And the deepest question, what does it do to us to share our homes with machines in our own image? We built them human shape to fit our world. Our world may change to fit them. A century ago, the tractor did not just change farming. It changed where people lived and what work meant. Historians still debate whether it destroyed jobs or created the modern world. It did both. The first humanoid you meet may not impress you. Slow, cautious, a little awkward. Remember the first automobiles. Slower than horses, breaking down on every road. Then remember what came next. The humanoid robot is that kind of machine. Not a better tool, but a new kind of worker. They are still clumsy. They still fail. The demos still flatter them. But they are here. They are working and they are learning. All of them, all at once, every single night. The machines are coming for every job. The only question is what we will do with everything they give back.

Assistir ao vídeo original no YouTube

Assuntos: Robótica e humanoides, Robótica e humanoides. Publicado no YouTube em 2026-10-18.

Seleção automática por assunto e idioma. O vídeo pertence ao canal de origem.