Combining Learned and Structured Knowledge for Robot Planning
Abstract: Machine learning is transforming robots' ability to perform tasks in diverse, real-world scenarios, but state-of-the-art learning methods often require extensive data and computational resources, limiting their accessibility to many communities. My work focuses on planning algorithms that combine the adaptability of machine learning with the efficiency and reliability of classical robotics methods. In this talk, we will build conceptual models for combining physics-based reasoning in classical robotics with tools from machine learning to enable planning in complex scenarios, such as plant watering. We will also explore strategies to solve real-world tasks using planning despite the inaccurate or incomplete predictive models available to us. I will discuss how these ideas inform how we can prepare each other for adaptive problem-solving in diverse contexts
Bio: Alex LaGrassa (they/them) is a PhD candidate in the Robotics Institute at Carnegie Mellon University researching methods that combine machine learning with classical robotics by quantifying then expanding robot capabilities. They develop algorithms that equip robots with the intelligence to manipulate challenging but common deformable objects such as plants, cables, and liquid with limited access to data and computational resources. Alex is also passionate about inclusive STEM education so all communities contribute to shaping technological development. They enjoy designing hands-on robotics teaching opportunities that invite students to participate fully in the learning process and contribute their unique perspectives.