Covert State Discovery and Multi-Agent Reinforcement Learning for Human-Autonomy Teaming
Abstract
In the proposed NICOP, Professor Lin and his team aim to develop a Multi-agent reinforcement learning (MARL) scheme that can be applied to the hybrid team of humans and autonomous agents. MARL is a relatively recent concept in machine learning that are being actively investigated by a number of research teams including those at Facebook AI Research and Microsoft Research. Some of the MARL applications have transitioned to the practical domains such as cyberattack countermeasures and robot control. There is a possibility that MARL frameworks will be able to address the distributed, cooperative, or hierarchical control problems in heterogeneous teams such as autonomous agent swarms and human-machine teams. However, in order to address such problems, MARL has to be able to first learn to deal with the difference in autonomy level, response time, and learning rate among the team members. The PI and co-PIs plan to tackle this issue by developing a covert state discovery algorithm that extracts the hidden covert-states transition diagram (COSTD) of the agents.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 23, 2019
- Source ID
- N629091912058
Entities
People
- Chin-Teng Lin
Organizations
- Office of Naval Research
- United States Navy
- University of Technology Sydney