A Systematic Review and Taxonomy of Human-Agent Teaming Testbeds.

Area of Science:

• Human-Computer Interaction
• Artificial Intelligence
• Team Dynamics

Background:

• Growing interest in human-agent teams (HATs) has led to diverse research testbeds.
• A lack of comprehensive understanding of existing HAT testbeds' capabilities and limitations exists.

Purpose of the Study:

• Develop a taxonomy for classifying human-agent teams (HATs).
• Conduct a literature review of existing HAT testbeds using the developed taxonomy.
• Analyze the capabilities and limitations of current HAT testbeds.

Main Methods:

• Modified an existing human team classification framework to create a ten-attribute HAT taxonomy.
• Analyzed 103 HAT testbeds from 235 empirical studies using the taxonomy.
• Performed frequency analyses on coded testbed attributes.

Main Results:

• Most testbeds feature single human participants with subordinate agents.
• Leadership is typically designated (human-led or none), with communication dynamics needing exploration for larger teams.
• Existing testbeds predominantly focus on long-term teams, neglecting ad hoc team dynamics.

Conclusions:

• Further research is vital for diverse team attributes like composition, leadership, and communication in HATs.
• Understanding complex team dynamics will enable the design of more effective human-agent teams.
• Findings guide future research directions for developing and utilizing HAT testbeds.