Have robots learned to fold laundry and make coffee after large models were applied?

Right now, we're witnessing large language models continuously giving rise to new capabilities, with each day bringing changes almost too numerous to take in. Google's chain-of-thought technology has dramatically enhanced the reasoning abilities of large models, while GPT-4's Reflexion goes a step further, endowing them with self-reflection. As we approach a future infinitely close to AGI, how can we translate these models into broader real-world applications? The integration of large models with robotics may open up a path to even greater possibilities.

This time, we've invited Yunzhu Li, an outstanding young researcher in robotics, machine learning, and computer vision. Li graduated from the Department of Computer Science at Peking University and earned his PhD from MIT. He is currently a postdoc at the Stanford Vision and Learning Lab (SVL), working with Fei-Fei Li and Jiajun Wu, and will join the University of Illinois Urbana-Champaign this fall. Back in 2019, a tactile glove developed by Li's team made frequent appearances in Chinese media, and his talks on TechBeat about robotic manipulation, physical interaction, and multimodal perception also received enthusiastic responses.

In this episode, you'll hear ZhenFund partner Emma Yin and ZhenFund investment manager Yuan Meng in conversation with Yunzhu Li, assistant professor in the Department of Computer Science at the University of Illinois Urbana-Champaign: How far has robotics development progressed? How are large models helping the field of robotics? What specific applications or scenarios can multimodal capabilities be deployed in? And how far are we from robots being part of everyday life?

Hosts: Emma Yin, ZhenFund Partner; Yuan Meng, ZhenFund Investment Manager

Guest: Yunzhu Li, Assistant Professor, Department of Computer Science, University of Illinois Urbana-Champaign

Timeline

2:55 What's changed in Silicon Valley this year

6:15 A robot is something with a physical embodiment that can interact with the environment

7:53 The pick-and-place problem in robotics: Amazon's picking challenges and progress across application scenarios

13:34 How large models help robots define and decompose tasks, expanding the boundaries of understanding

21:24 What is intuitive physics? What are the steps in robot-environment interaction?

29:12 The tactile glove and tactile carpet developed by Li's team

37:45 How far are robots from ordinary people's daily lives?

42:05 The 1,000 tasks ordinary people want robots to accomplish

45:02 Advice for young people who want to pursue AI research

Related Materials

YunZhu Li

Yunzhu Li - TechBeat Talk

MIT's $10 AI "Thanos Glove" Is Here! 548 Sensors, Identifies Objects With a Light Touch

Production

Post-production: Chong'er

Contact Us

WeChat Official Account: ZhenFund (ID: zhenfund)

Listen on: Xiaoyuzhou | Apple Podcast | Ximalaya

Email: media@zhenfund.com

If you have any suggestions or expectations for the show, we welcome your feedback and interaction in the comments~