Doctoral Thesis: Learning State and Action Abstractions for Effective and Efficient Planning

Rohan Chitnis

Abstract:

An autonomous agent should make good decisions quickly. These two considerations — effectiveness and efficiency — are especially important, and often competing, when an agent is planning to make decisions sequentially in long-horizon tasks. Unfortunately, planning directly in the state and action spaces of a task is highly intractable for many tasks of interest. Abstractions offer a mechanism to overcome this intractability, allowing the agent to reason at a higher level about the most salient aspects of a task. In this thesis, we develop novel frameworks for learning state and action abstractions that are optimized for both effective and efficient planning. Most generally, state and action abstractions are arbitrary transformations of the state and action spaces of the given planning problem; we focus on task-specific abstractions that leverage the structure of a given task (or family of tasks) to make planning efficient. In this talk, we show how to learn neuro-symbolic abstractions for bilevel planning, and demonstrate in robotic planning tasks that the methods we present optimize a tradeoff between planning effectively and planning efficiently.

Details

• Date: Monday, April 25
• Time: 10:30 am - 12:00 pm
• Location: 32-G449 (Patil/Kiva Seminar Room)

Additional Location Details:

Thesis Supervisors: Profs. Leslie Kaelbling and Tomas Lozano-Perez