Neuro-Symbolic Learning for Context-Aware Real-Time Human-Agent Interaction
Abstract
Robotic agents have been developed to perform a number of tasks under human command and sometimes independently as well. Human-agent teams can extend the capabilities of both humans and robots. Multi-Agent Reinforcement Learning (MARL) is a natural approach for such teams. The interaction within a team composed of human and robot agents considered so far ignores the role of physical context in MARL. Here, we present a distributed online MARL in such a scenario incorporating wireless communication and physical movement supplemented with intra-team communication. We present the system, its mathematical analysis, and some initial experimental results herein
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2024
- Accession Number
- AD1230960
Entities
People
- Julian De Gortari Briseno
- Mani Srivastava
- Vinod K. Mishra
Organizations
- United States Army Research Laboratory
- University of California