"GEN-1 achieves 99 percent success rates on delicate mechanical tasks such as folding boxes and servicing robot vacuums, operating at three times the speed of GEN-0."
"The system can respond to disruptions by improvising new moves and connecting ideas from different places to solve new problems."
"Generalist has collected over half a million hours of physical interaction data using 'data hands' to train its physical model."
"GEN-1 can adapt its pretraining to specific robotic tasks after only about an hour, demonstrating significant efficiency in learning."
Generalist has introduced GEN-1, a physical AI system that achieves production-level success rates in various physical skills previously requiring human dexterity. The system can adapt to disruptions and improvise solutions. Building on the GEN-0 model, GEN-1 utilizes over half a million hours of data collected through wearable pincers to train its capabilities. It demonstrates a 99 percent success rate in delicate tasks and operates three times faster than its predecessor, adapting quickly to specific robotic tasks after minimal training time.
Read at Ars Technica
Unable to calculate read time
Collection
[
|
...
]