A few months ago we said the next breakthrough in robotics won’t come from copying more demonstrations 💡
Robots need to understand the job: what must be true before an action, what can fail, how to recover
Amazing to see the field recognising this. We’ve been building on it
Not one π0 weight changed.
Precision fitting on a real robot: 0/10 → 9/10.
Early results.
Part 1 of 3: more capability from the models you already have. 🧵
These are early results. More details are in the post:
https://t.co/yVYI4YcmYI
Coming next:
Part 2: the cost of getting work done.
Part 3: from intelligent models to intelligent agents.
Same pattern with world models.
LeWM by @amilabs with our Work Model on top:
PushT 90% → 96%
OGB-Cube 70% → 84%
Where LeWM alone drops the cube, the same model with Motoniq reaches the goal.
A few months ago we said the next breakthrough in robotics won’t come from copying more demonstrations 💡
Robots need to understand the job: what must be true before an action, what can fail, how to recover
Amazing to see the field recognising this. We’ve been building on it
I have so much fun writing this position with some of the most amaaazing people in robotics!
Have a look at it here: https://t.co/zM3NBtobkx
#AI#MachineLearning#Robotics