Constructing Abstraction Hierarchies for Robust, Real-Time Control
Abstract
This project primarily focused on the theoretical principles underlying which high-level actions an agent should build, and data-efficient algorithms for learning those high-level actions from interaction with an agent's environment. The projected funded a single PhD student for three years, and resulted in 5 publications at top-tier, highly-refereed international conferences, and 3 additional publications either in preparation or currently under review. The report describes these research results and draws appropriate conclusions.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 22, 2020
- Accession Number
- AD1103140
Entities
People
- George Konidaris
Organizations
- Brown University