Monday / CS Seminar

March 17th / 12:00 PM

Hybrid / Mudd 3514

Speaker

Ruohan Zhang, Stanford Vision

Talk Title

Toward Human-Centered Embodied AI

Abstract

My research focuses on developing human-centered, human-inspired, and human-compatible embodied AI, structured around three pillars: human tasks, human data, and human interfaces. To systematically study intelligent behaviors (both human and artificial) in natural tasks and environments, we develop BEHAVIOR, a comprehensive simulation toolkit that supports 1,000 essential everyday tasks identified through large-scale survey studies. BEHAVIOR instantiates these tasks in interactive and ecologically valid virtual environments utilizing the state-of-the-art physics simulator. My approach to achieving human-level performance in these tasks requires embodied AI systems to learn from multimodal human data, such as body motion, eye movements, language, and reward signals. To enable such learning, we design novel interfaces that capture diverse human behavior data at scale, alongside algorithms that allow embodied agents to learn from such data effectively. To augment human capabilities and overcome their physical limitations (e.g., for patients with motor disabilities), we develop intelligent brain-robot interfaces, leveraging intracortical arrays and electroencephalogram (EEG), that can assist individuals in performing everyday tasks. For future work, we hope the infrastructure we have built can serve as powerful open-source tools that inspire and support a new generation of human-centered AI research, with the goals of understanding human brains and behaviors in natural tasks and environments, developing human-inspired and human-compatible AI solutions, and augmenting human capabilities with embodied AI.

Biography

Ruohan Zhang is a postdoctoral researcher at the Stanford Vision and Learning Lab, Institute for Human-Centered Artificial Intelligence (HAI), and Wu Tsai Human Performance Alliance, advised by Fei-Fei Li and Jiajun Wu. His research focuses on embodied AI, human-robot interaction, brain-computer interfaces, and cognitive science. He earned his Ph.D. in Computer Science from the University of Texas at Austin, advised by Dana Ballard and Mary Hayhoe. He has received several paper awards or nominations from robotics conferences and workshops. He has served as a program committee member, reviewer, or editor for conferences and journals in machine learning, robotics, and cognitive science, including NeurIPS, ICCV, ECCV, CoRL, RSS, ICRA, IROS, CogSci, Frontiers, and the Journal of Vision.

Research/Interest Areas

robotics; human-robot interaction; cognitive science; neuroscience

---

Zoom: https://northwestern.zoom.us/j/97629411184?pwd=t01DJrwaREICRbCoHbi3bhvlMVFvz4.1

Panopto: https://northwestern.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=e1faba26-2fd5-45ee-b016-b29e01655df1