|DATE||January 19 (Tue), 2021|
|TITLE||Embedded AI: Human-centered robotics|
Human-centered robotics is the development of highly capable autonomous robotic collaborators designed to help humans in real environments across homes, factories, and outdoors. In recent years, machine learning has advanced robotic perception or control. Yet for state-of-the-art robots, such tasks prove challenging without humans' supervision to perform complex tasks with humans. The robots also require a means of safely executing requested tasks while inferring complete task plans from ambiguous orders. My research focuses on learning manipulation capabilities that enable robots to effectively assist people in various environments leveraging natural language understanding, multimodal perception, and representation learning. In this talk, I will introduce three core capabilities with task representations that advance collaborative manipulation in human environments: skill learning for complex tasks, safety monitoring for assistive manipulation, and planning for human-robot team operations. I will show how to provide highly scalable and reliable assistance when situated in novel environments.