China Just Dropped Self Evolving AI Robots With Real Human Physical Intuition

Ajouté : 2026-05-31

[00:00:00]For a long time, the idea of intelligent robots has been framed the wrong way.

[00:00:06]We've been focused on intelligence.

[00:00:08]Faster processing, better models, more data.

[00:00:13]But intelligence alone was never the real barrier.

[00:00:16]The real barrier was understanding the physical world.

[00:00:22]Because there's a difference between knowing something and experiencing it.

[00:00:27]A machine can calculate trajectories, simulate physics, and predict outcomes.

[00:00:33]But that's not the same as feeling instability, adjusting balance, or learning from failure in real time.

[00:00:41]Humans don't learn from perfect information.

[00:00:44]>> [music] >> We learn from mistakes, from trial and error, from things going wrong, from moments where we misjudge something, and correct ourselves. And for decades, robots couldn't do that. They could follow instructions perfectly until something unexpected happened.

[00:01:02]And then they failed.

[00:01:04]Not because they weren't intelligent, but because they didn't truly understand the world they were operating in.

[00:01:10][music] That's what's starting to change.

[00:01:13]Not slowly, not gradually, but through multiple breakthroughs happening at the same time.

[00:01:20]A robot that can physically adapt to force in real time.

[00:01:24]A humanoid cheap enough to exist outside of labs.

[00:01:29]And a learning system that allows machines to develop something close to intuition.

[00:01:34]Individually, each of these is impressive.

[00:01:37]Together, they signal a shift, not just in what robots can do, but in what they are becoming.

[00:01:46]G1 Robot control, not strength.

[00:01:52]Let's start with something that looks simple, but it isn't.

[00:01:56]A humanoid robot pulling a car.

[00:01:59]At first glance, it feels like a strength demonstration, but if you look closely, that's not what's happening.

[00:02:06]Because strength isn't the impressive part.

[00:02:09]Control is.

[00:02:11]The robot doesn't just pull the car forward, it manages the interaction between its body and the environment.

[00:02:18]It leans backward to generate force.

[00:02:21]It adjusts its footing continuously. It shifts its center of gravity with every step. And most importantly, it reacts in real time.

[00:02:31]Because the moment you attach a heavy object to something moving, everything becomes unstable. The load shifts.

[00:02:40]Friction changes.

[00:02:42]Balance becomes unpredictable.

[00:02:45]And yet the robot compensates for all of it.

[00:02:49]Not with pre-programmed movements, but with continuous feedback. This is what matters.

[00:02:56]Because real environments are not controlled environments. They're messy, unpredictable, full of variables that can't be perfectly modeled in advance.

[00:03:06]And for a robot to operate in that kind of space, [music] it needs to adapt.

[00:03:12]The problem of balance.

[00:03:15]Balance sounds simple, but it's one of the hardest problems in robotics.

[00:03:20]Humans do it automatically.

[00:03:22]You lean slightly without thinking.

[00:03:25]You catch yourself before falling.

[00:03:27]You adjust to uneven surfaces without conscious effort.

[00:03:32]All of that happens below awareness.

[00:03:34]But for a machine, every one of those adjustments has to be calculated, measured, predicted, corrected, and even then, it's difficult.

[00:03:45]Because the world isn't perfectly predictable.

[00:03:48]A small shift in weight can change everything.

[00:03:52]A slight incline can affect movement. An unexpected force can destabilize the entire system.

[00:03:59]For years, robots struggled with this.

[00:04:02]They could move well in controlled environments.

[00:04:05]Flat floors, stable conditions.

[00:04:08]But introduce variability and things broke down.

[00:04:13]That's why this kind of demonstration matters.

[00:04:16]Not because of what the robot lifted, but because of how it maintained control while doing it.

[00:04:22]Boomy.

[00:04:23]The price revolution.

[00:04:26]Now, while some companies are pushing the limits of performance, others are changing something even more fundamental.

[00:04:34]Price.

[00:04:35]Because for years, humanoid robots have existed, but they've been inaccessible.

[00:04:41]Expensive systems confined to labs, research centers, and large corporations.

[00:04:47]That's beginning to shift.

[00:04:50]A humanoid robot was introduced at a price point that changes the equation entirely. Around $1,300.

[00:04:58]That's not a research budget. That's consumer territory.

[00:05:02]And while it's smaller, simpler, less powerful than high-end machines, it still does something important. It exists outside the lab. It walks, [music] it balances, it interacts, and it does so at a cost that makes it reachable.

[00:05:19]Why cost matters more than capability.

[00:05:22]Here's something history has proven repeatedly.

[00:05:26]Technology doesn't transform the world when it becomes perfect. It transforms the world when it becomes accessible.

[00:05:32][music] The first computers weren't revolutionary for most people.

[00:05:37]They were expensive, limited, and out of reach.

[00:05:41]But as costs dropped, adoption increased.

[00:05:45]And once adoption increased, innovation accelerated. The same thing happened with smartphones. At first, they were niche.

[00:05:54]Then they became a widespread and eventually an essential.

[00:05:58]Robotics may follow the same pattern because a smaller, less powerful robot that people can actually own will shape the future more than a perfect system that remains inaccessible.

[00:06:13]The real limitation, learning.

[00:06:17]But even with better hardware and lower costs, there has always been a limitation, learning. Not just processing information, not just identifying patterns, but understanding how the world works.

[00:06:34]Because most AI systems today are passive. They observe, they analyze, they predict, but they don't truly understand cause and effect. They know what usually happens, but not why it happens.

[00:06:51]A different approach to intelligence.

[00:06:54]A new approach is trying to change that.

[00:06:57]Instead of relying only on data, it introduces interaction. The system makes predictions, tests them, evaluates the outcome, adjusts its understanding, >> [music] >> and then repeats the process again and again. This loop, trial, error, correction, is how humans develop intuition.

[00:07:21]You don't learn balance from reading about it.

[00:07:23]You learn it by almost falling.

[00:07:26]You don't understand fragility from theory.

[00:07:29]You understand it by breaking something once and remembering it.

[00:07:34]This kind of learning is messy, [music] imperfect, but powerful.

[00:07:38]And for the first time, machines are starting to use it. What comes next?

[00:07:46]We're not at the final stage.

[00:07:48]Robots still struggle with many things.

[00:07:51]Fine motor control, delicate manipulation, complex decision-making.

[00:07:58]Tasks that are simple for humans remain difficult for machines, but the direction is clear and the pace is increasing, which means the gap is closing.

[00:08:08]Here's the part most people underestimate.

[00:08:11]None of these developments on their own change everything.

[00:08:14]But together, they create momentum.

[00:08:18]One improves movement.

[00:08:20]One improves accessibility.

[00:08:23]One improves learning. And when those elements combine, progress accelerates. That's where we are right now.

[00:08:31]Not at the end of the story, but at the point where things start moving faster than expected.