Choosing the right object is part of the task.
The video shows what gets used and what stays put.
A crowded table can make that choice harder.
Robots need to learn which objects are relevant 🦾
Keep recording until the job is actually done.
That includes the final check of the result.
After closing a lid, show that it stays shut.
The clip should show whether the task succeeded 🦾
Future robot trainers may already be at work.
Cooks and technicians repeat skilled tasks every day.
Egocentric video captures how those jobs really happen.
Their experience can become training data for robots.
The workers doing the task can help teach it to robots 🦾
Experts and beginners can perform the same task.
Experts pause less and follow a more efficient sequence.
Beginners reveal where the task becomes difficult.
The best way to capture data is to act naturally.
Because the most important thing is to understand how humans behave.
Packaging is a hidden test for robot hands.
Caps twist. Tape peels. Zippers pull. Seals tear.
Each one demands a different grip, direction and force.
Egocentric video captures how people handle each one.
Every package adds another useful manipulation task 🦾
RLWRLD's HuRo turns human videos into training data.
The dataset reaches about 630K robotized episodes.
As the data scaled, out-of-distribution success rose from 35% to 72%.
More human video, stronger robots.
Egocentric data keeps proving its value 🦾
https://t.co/5qQeQ47BaR
🎉 Excited to share HuRo, accepted to CoRL 2026!
Can robotized human videos provide scalable robot learning supervision?
HuRo turns human videos into robot aligned observations and actions.
In VLA pretraining, OOD completion rises from 34.9% to 72.2%.
🧵 More in replies.
Humans rarely complete tasks with textbook motions
A jar gets braced against a table before it is opened
A bag gets held with an elbow while both hands work
There are things even humans are unaware of
That is why egocentric video is so important for capturing human behavior
HumanEgo won Best Paper at IROS 2026 WORLDS.
It teaches robots from minutes of human egocentric video.
Human video is rich in quantity but poor in quality.
Smarter representations matter as much as more data.
They open-sourced everything 🦾
https://t.co/KE0DfgaOqT
Season 7 rewards are live 💰
200,000 $VADER has been distributed to eligible contributors.
Season 8 is live now. Complete tasks, contribute data, earn $VADER 🦾