LCSR Seminar: Meghan Booker

Description

Meghan Booker, a postdoctoral fellow for the robotics group in the Research and Exploratory Development Department at the Johns Hopkins University Applied Physics Lab, will give a talk titled "Task-Driven Perception and Control for Robust and Efficient Autonomy" for the Laboratory for Computational Sensing + Robotics.

Abstract:

In order for robots to operate reliably in new, complex, and uncertain environments, robots need to be able to effectively use the information provided to them about their environment. This becomes particularly important if the robot is using high-dimensional sensor observations (e.g., RGB-D cameras for visuomotor control) or an information-rich map (e.g., 3D scene graphs for task planning) as the robot needs to be able to make real-time decisions and be robust to perception errors. Through the lens of task-driven perception and control, we can obtain representations and corresponding control policies that are computationally efficient and robust specifically to task-irrelevant distractors. In this talk, I will highlight several algorithmic approaches for obtaining such task-driven control policies in the contexts of model-free reinforcement learning, model-based control, and large language model–based planning.

Bio:

Meghan Booker received a Ph.D. in mechanical and aerospace engineering from Princeton University where she was a Gordon Y. S. Wu fellow, and in 2018, she received a B.S. in electrical and computer engineering from The Ohio State University. Booker's research focuses on developing task-driven algorithms for reliable and adaptive robot deployment.