Inverse Reinforcement Learning with High-Level Task Information (Year 1)
Abstract
This report describes research to discover means by which future Army robotic systems could learn new behaviors from human teammates while still verifiably meeting system and mission specifications. Within the first year, research was conducted on learning jointly from human demonstrations and specifications given in linear temporal logic, under conditions of partial observability. The research demonstrated success within a simple grid world environment. Research is ongoing to extend this progress to more complicated environments, such as that of the US Army Combat Capabilities Development Command Army Research Laboratory's autonomy stack Unity environments.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2021
- Accession Number
- AD1149439
Entities
People
- Craig Lennon
Organizations
- United States Army