10:24
09:52
11:01
10:52
01:59
01:33
10:24
09:52
11:01
10:52
01:59
01:33
10:24
09:52
11:01
10:52
01:59
01:33
10:24
09:52
11:01
10:52
01:59
01:33
NVIDIA Research has unveiled Eureka, an AI agent powered by GPT-4, which has successfully trained robots to perform a wide range of tasks using reward algorithms.
One impressive feat is teaching a robotic hand to execute pen spinning tricks as skillfully as a human, if not better. Eureka has also trained robots to perform nearly 30 different tasks, such as opening drawers, using scissors, and catching balls.
These AI-based reward programs were found to be 80% more effective than those crafted by human experts, leading to a 50% improvement in the robots' performance. Furthermore, Eureka can self-evaluate and adjust reward functions based on training results.
NVIDIA Research has made its Eureka algorithms available for others to explore and utilize in NVIDIA Isaac Gym, a physics simulation reference application for reinforcement learning research. The concept of robot learning is gaining momentum in AI research.