Virtual Robot Duck Trained in Hours With AI, Hinting at Faster Robot Development

At a glance

A writer used OpenAI's GPT-6 Sol to train a virtual Microduck robot to walk around a room and find bengal cats. The job took a few hours, which used to take weeks.

Key points

1. 1

   Microduck is a small robot made by Pollen Robotics and Hugging Face.

2. 2

   Training used Nvidia's Isaac Sim, and the AI watched videos of the robot to improve its training.

3. 3

   The robot learned to walk without falling over, but its attempt at a moonwalk ended as a clumsy backward jog.

4. 4

   UC Berkeley researchers said the latest AI built a detailed 3D robot simulation in about an hour.

Why it matters

As AI helps train robots, building them is becoming easier and cheaper for more people.

Also reported by

• Martin Cid Magazine로봇 공학의 미래: 기자 윌 나이트, AI로 가상 Microduck을 몇 시간 만에 훈련해 고양이 찾게 해