madrocks

Crypto is future and future is crypto

[madrocks @thetrader996]

@RemotejobWtessy I am interested

[madrocks @thetrader996]

100,000 robot experiences sound impressive.

But what if the robot did the same task 100,000 times?

That’s the problem with robot training data.

A robot needs to see the same skill in different situations:

Different objects.
Different positions.
Different environments.
Different ways of completing the task.

Because the real world doesn’t repeat itself perfectly.

This is why I think **data diversity** may matter more than simply chasing bigger numbers.

@AxisRobotics is building around this idea — generating varied tasks and environments through simulation and distributed data collection.

More data is good.

But **more diverse data is what helps robots generalize.**

That could be one of the biggest challenges in Physical AI.

Explore Axis Hub:
https://t.co/vh5sVpaEiP

#AxisRobotics #PhysicalAI #Robotics #AI

[madrocks @thetrader996]

Humanoid robots aren’t just about making robots look like humans.

The real breakthrough is making robots capable of working in the same environments humans already built.

Factories.
Warehouses.
Homes.
Hospitals.

No need to redesign the entire world for machines.

That’s why I’m watching @axisroboticsai closely.

The future isn’t humans vs robots.

It’s humans + robots working together. 🤖

#AxisRobotics #Kaito #Robotics #AI

Join the Axis Robotics ecosystem:
https://t.co/L3SPprdhQU

[madrocks @thetrader996]

That’s a fair point. The real challenge isn’t just collecting corrections—it’s making sure the robot actually generalizes beyond the failures it has already seen.

A strong system should keep training and evaluation data strictly separated, with unseen operators/tasks used to test whether the policy truly improved.

Great callout 👌

[madrocks @thetrader996]

What if a robot could learn from its mistakes?

Imagine a robot trying to pick up a cup.

It gets close…

but misses the grasp.

Instead of starting over, a human can step in and correct the movement.

That correction becomes training data.

Robot tries → human corrects → data is collected → policy improves → robot tries again.

That’s what caught my attention about @AxisRobotics and its V2 approach.

The interesting part isn't just collecting more robot data.

It’s turning failures into useful learning signals.

If robots can learn from their mistakes at scale, Physical AI could improve much faster.

Explore Axis Hub:
https://t.co/vh5sVpaEiP

#AxisRobotics #PhysicalAI #Robotics #AI

[madrocks @thetrader996]

Definately

[madrocks @thetrader996]

Why can’t we train robots the same way we train ChatGPT?

Because a robot doesn’t just need to understand the world.

It needs to act in it.

AI models can learn from billions of words, images and videos.

But a robot needs examples of:

→ what it sees
→ how it moves
→ what action it takes
→ whether the task succeeds

That sequence of actions is valuable training data.

And collecting it with real robots can be slow, expensive and difficult to scale.

That’s what makes @AxisRobotics interesting to me.

Axis is exploring scalable robot-data collection through simulation and human demonstrations.

The bigger idea:

Smarter robots need more experience.
More experience needs better data.

I think the companies that solve this data problem could play a major role in the Physical AI stack.

Explore Axis Robotics:
https://t.co/vh5sVpaEiP

#AxisRobotics #PhysicalAI #Robotics #AI

[madrocks @thetrader996]

Most people think the bottleneck for Physical AI is better models.

I think the bigger bottleneck is data.

A robot can have an impressive model, but if it doesn't have enough diverse examples of how humans actually interact with the physical world, intelligence has nothing to learn from.

That's where @axisrobotics gets interesting.

Axis is building a data engine where people can generate robotic trajectories through interactive tasks, turning human interaction into training data for Physical AI.

And there's an important idea behind this:

More diverse data → better learning → better robot policies → more useful robots → more data.

That's a compounding loop.

Axis has already shown that community-generated trajectories can improve robot-policy performance: 50K+ trajectories lifted π0.5 on LIBERO-Plus from 83.9% to 88.8%.

We're still early in Physical AI.

The companies building the data infrastructure, not just the models, could end up being extremely important.

I'm watching @axisrobotics closely. 🤖

@KaitoAI
#AxisRobotics #PhysicalAI

[desi mojito @desimojito]

This is my story

[madrocks @thetrader996]

@KashifK24401071 @MrTiwaree Q bich me economy ko lathe ho directly bolo na ki bjp hate muslims

[madrocks @thetrader996]

@MrTiwaree We stand with modi with more conviction

[madrocks @thetrader996]

@idfcwau Chutiye abhi south and north shuru matt Karlena ham south vale respect everyone varana ham bolne pe utar aye tho soch north ke logo ke bare me kya kaya bol sakte hai we don't divide india by south and north

[madrocks @thetrader996]

@GemsOfWood Bro don't even try to compair any movie with bahubali

[madrocks @thetrader996]

@ripper_dhruvaa Chutiye the person in left side is from Karnataka, and we don't eat beef

[madrocks @thetrader996]

@vijaythottathil Kerala saar 100% literacy saar

[madrocks @thetrader996]

@isthatabdullah7 @CMShehbaz Tum log hamesha apne hi sapne me hote hoo

[madrocks @thetrader996]

@Shivayaaah I too like dhurandar but as a Kannadiga I know yash He is different

[madrocks @thetrader996]

Hey @grok tell me how many colours are there

[madrocks @thetrader996]

@RVCJKannada Mark

[madrocks @thetrader996]

@RajatMemes_ Didi