Following Rocky's Journey
Fascinated by the remarkable speed at which humans learn, Rocky Duan began his AI research journey at UC Berkeley, specializing in robotics motion planning research. After graduating with three majors in three years, he continued on to UC Berkeley’s PhD program where he completed his doctoral studies in a swift 2.5 years.
In 2015, Rocky joined OpenAI as one of its earliest team members while working on his PhD. His pioneering work on "learning to learn" marked a significant industry advancement in the field, allowing robots to acquire new skills with minimal trial and error.
While at OpenAI, Rocky developed techniques that significantly improved the efficiency of robotic learning. These methods drew from past experiences, enabling robots to adapt their learning strategies automatically, a notable departure from the traditional approach of teaching robots from scratch. This approach laid the foundation for Transformers, a fundamental component of well-known language and vision models, including ChatGPT and DALL·E.
Motivated by a desire to translate fundamental research into real-world impact, Rocky co-founded Covariant in 2017. Alongside AI researchers Pieter Abbeel, Peter Chen, and Tianhao Zhang, they have been at the forefront of AI Robotics, creating innovative solutions that automate labor-intensive tasks within the warehousing industry, a critical driver of global e-commerce.
In his role as Chief Technology Officer, Rocky leads the development of the Covariant Brain, our universal AI platform. This platform empowers hundreds of robots worldwide and holds the distinction of being the first foundation model to be deployed in the physical world.
Get the full scoop:
- 02:44 About Covariant
- 04:00 Warehouse pick and place
- 7:59 Real-world deployments
- 10:02 Fully integrated solutions (vs. just the AI)
- 13:04 Covariant's system in-depth
- 22:08 Role of foundation models
- 30:17 Robotics' current frontier
- 39:01 AI Robotics commercial applications
- 46:30 Leaving OpenAI to found Covariant
- 50:44 Original learning to imitate and learning to reinforcement learn
- 59:24 Rocky's tips for productivity
