|Speaker:||Dr. Hao Zhang, Colorado School of Mines|
|When:||Friday, January 12, 2018, 11:00am - 12:00pm|
|Where:||Research Hall, Room 420|
A significant demand of modern robotics has been for robots that can work collaboratively alongside humans for long periods of time. Yet, robot learning methods, while allowing robots to effectively work with single individuals toward short-term tasks, are poorly suited for robots to collaborate with a human team for long-term autonomous operation from hours to lifetime in novel environments. Long-term collaborative autonomy is indeed ubiquitous in real life, especially in defense applications in adversarial situations. In this talk, I will describe two recent and interrelated advances in robot learning aimed at enabling long-term collaborative autonomy. The first focuses on learning representations of the environment with long-term appearance changes for loop closure detection in robot localization and mapping. The second is teamwork modeling for team intent and individual behavior recognition that is useful for planning and interaction. In contrast to the prevalent 'Big Data' approach, we focus on 'Small Data' regimes that can learn from a minimal set of training instances. I will also present our recent in-sights on how to leverage adversarial artificial intelligence techniques to link learning with security to enable robot adaptation in adversarial settings.
Hao Zhang is an Assistant Professor of Computer Science at the Colorado School of Mines. He is the director of the Mine's Human-Centered Robotics Laboratory (http://hcr.mines.edu). Dr. Zhang completed his PhD at the University of Tennessee, working on distributed algorithms to enable peer-to-peer human-robot teaming, where he received the prestigious Chancellor's Fellowship. His current research interests broadly lie in the fields of robotics, machine learning, artificial intelligence, and augmented reality. His core research is in regularized optimization algorithms and sparsity models that enable long-term and collaborative autonomous operation of robotic systems, particularly in adversarial or unknown environments. Dr. Zhang has extensive experience in designing real-time multimodal learning systems for robots, and is interested in fundamental understanding of statistical learning with sparsity that can be used to represent the state of the world and human-robot teams. His research uses heterogeneous robots equipped with a variety of sensors to advance and validate algorithms that are useful for enabling long-term collaborative autonomy.