Generalist announces GEN-1 AI system for robot hands and factory tasks

In short: Generalist says its new GEN-1 robotics AI helps robots do delicate, repetitive tasks with high success rates and adapt when something goes wrong.

What happened

Generalist, a company that trains robots with machine learning (software that learns from examples instead of following fixed instructions), announced a new system called GEN-1. The company says GEN-1 can handle a wide range of hands-on tasks that usually need human dexterity.

One problem with training robots is data. Language AIs can learn from huge amounts of text online, but there is no equivalent library showing, step by step, how hands move objects. Generalist says it collected this kind of information using “data hands,” wearable pincers that record small finger and hand movements plus video while a person does real tasks.

Generalist claims it now has more than half a million hours of recorded human actions to train GEN-1. In company demos, the robot can do tasks like putting cash into a wallet, folding laundry, and sorting auto parts. Generalist says GEN-1 reaches about a 99 percent success rate on repetitive but delicate work like folding boxes, packing phones, and servicing robot vacuums, and it runs roughly three times faster than its earlier GEN-0 model.

Why it matters

Robots often fail when something unexpected happens, like an object slipping or bending. Generalist says GEN-1 can recover by adjusting its grip and trying new moves, a bit like a person improvising when a sock twists while folding laundry. If these claims hold up outside of demos, it could make robots more useful in places like warehouses and factories, where small mistakes can stop a whole workflow.

Source: Arstechnica