
Professor Yuke Zhu delivers a presentation on humanoid robotics, covering recent advancements, opportunities, and challenges in the field. He discusses the evolution of humanoid robots, the impact of AI and hardware development, and the potential for general-purpose robot autonomy. Zhu introduces projects like Okami and GrootM1, which leverage human video and large-scale data for robot learning. He also touches on the data pyramid concept, emphasizing the importance of web, synthetic, and real robot data, and explores strategies like co-training and generative AI tools for creating simulation environments. The presentation concludes with a discussion on the significance of humanoid local manipulation and the need for improved hardware and software co-design, followed by a Q&A session addressing safety concerns, the role of noise in training, evaluation methods, and the future of robotics.
Sign in to continue reading, translating and more.
Continue