NVIDIA has developed an AI model that helps robots learn complex movements like rotating a pen

By: Bohdan Kaminskyi | 21.10.2023, 16:58
NVIDIA has developed an AI model that helps robots learn complex movements like rotating a pen
NVIDIA

NVIDIA has unveiled Eureka, a GPT-4-based AI agent that trains robots to perform various tasks using reinforcement learning algorithms.

Here's What We Know

In particular, Eureka taught a robotic arm to rotate a pen as well as a human. The agent has also trained four-legged robots, manipulators and other mechanisms to open drawers, use scissors, catch balls and about 30 other skills.

According to NVIDIA, Eureka's trial-and-error reinforcement programmes are 80 percent more effective than those written by experts. As a result, robot performance has increased by more than 50 per cent.

Eureka also independently evaluates learning results and adjusts reinforcement algorithms.

NVIDIA has published a library of Eureka algorithms and invited other researchers to test them in the Isaac Gym simulator to model physical processes.

Source: NVIDIA