A robot getting a task right 95% of the time sounds impressive.

In a real deployment, it can also mean 1 failure in every 20 attempts.
Agility Robotics’ CTO Pras Velagapudi describes a useful way to think about the problem: robot telemetry doesn't just tell…

Dexterous robots don’t just need to see the world correctly.

They need to feel how the world pushes back.

Project SuperDex, a simulation platform built for dexterous manipulation, is now open source.

The platform is built around a problem that is easy…

For Physical AI, bad data isn’t just a cleanup problem.

Sometimes, the failure happens before the data even leaves the room.

When robots learn manipulation from human demonstrations, egocentric recordings need to capture the wearer’s hands consistently.…

Most 3D reconstruction starts with a simple question: what does this place look like?

Atlas adds a more powerful one: what would I see if I moved the camera there?

World Labs’ new foundation model for spatial intelligence can reconstruct real spaces from…

Scientific experiments can be bottlenecked by something surprisingly basic: the machines don't speak the same language.

In the brain-imaging setup shown here, a laser has to move and align precisely across multiple components. Each device may have its own…

A car used to move because we told it where to go.

XPENG is showing a different model of mobility: machines that can interpret the world around them and act accordingly.

In the clip, the progression is striking from vehicles that understand their surroundings…

Many of us want home robots. So what’s actually keeping them out of our living rooms?

It’s not just whether a robot can move.

A robot can learn to lift a plate, carry a cup, or reproduce a complex motion. But the same force that lets it handle an object…