Humans learn by watching. Robots should too.

Unfortunately, cooking is not our talent. So we teach our robots to cook for us.

See our robot open doors, water plants, assemble a box, and more by learning from watching humans. Using <1hr of robot data.

Skild Brain is robust to adversarial disturbances and transfers zero-shot to unseen homes.

Skild Brain is omni-bodied. Any robot, any task. See it load a dishwasher using a different form-factor.

Ever lost your AirPods? Here, a different embodiment assembles 10+ AirPods cases in a row, requiring extreme precision and dexterity.

Skild Brain is highly adaptable and reasons via in-context learning.

Blog → Careers → Follow @SkildAI for updates.

we live in the future

Indeed, grateful for @lightspeedvp's continued support!!

@lightspeedvp 🫡

🥳🥳🥳🥳🥳🥳🥳

🦾🧡

This is an insane breakthrough - huge congrats to the Skild team

@GMNDProtocol is building the eyes that validate what they watch

cool but how about non static tasks that require the robot to walk around aswell

Long-horizon generalization is less about clever architectures, and more about seeing enough diverse, real-world behavior over time.

@SkildAI, that's a concise point, and imitation learning is a great way to train robots, indeed.

amazing object recognition and fluidity of motion... even under duress!

Yes! And the next step is making that demonstration data reusable and accessible. One company’s teleoperation sessions = training data for the whole ecosystem. That’s how you scale imitation learning beyond single labs.

Would love to see examples of it learning purely from watching without any preset motions

That's cool! If robots can really learn by watching, then the bottleneck to robotics shifts from hard-coded control to data and imitation. Deployment can get much cheaper

how do you compare with FSD, as both being Video learner?

Does this mean the act of collecting egocentricdata from Asian countries is dead???

The gap between watching a video and actually feeling the torque is still huge. Interested to see how they bridge it.

I wonder how old are this abilities?

Nice! We got VLAs now 🦾😎💯

Soothing watching 👀

watch and learn

This is exactly why the best robotic designs use a humanoid design. It's much easier to teach a robot to do tasks when they can simply watch us and mirror what they see, limb or limb, joint for joint.

The birth of AI and robotics, along side God and Man.

great

the cheeks on that robot

Progress is advancing / @xmaquina 👀

Learning from watching and then trial &amp; error is definitely is the most scalable way forward that will make the most significant difference.

If a robot touches a hot stove what happens?

stove becomes cold?

before or after the robot melts or gets damaged?