ARTFEED — Contemporary Art Intelligence

Influence-Based Team Steering Framework for Zero-Shot Human-Machine Teaming

ai-technology · 2026-05-18

The newly introduced framework, Influence-Based Team Steering (IBTS), aims to overcome the challenges faced by zero-shot coordination (ZSC) in human-machine collaboration. Although ZSC attempts to mimic various partner groups to minimize the need for expensive human interaction data, it encounters difficulties as team sizes increase and communication deteriorates. IBTS promotes influence shaping to encourage agents to identify diverse and effective interaction strategies while guiding ongoing processes towards improved coordination. The framework was evaluated using Overcooked-AI in scenarios involving both two and three agents to examine its applicability beyond dyadic interactions. This research is available on arXiv under ID 2605.15400.

Key facts

  • arXiv ID: 2605.15400
  • Announce type: new
  • Proposes Influence-Based Team Steering (IBTS) framework
  • Addresses limitations of zero-shot coordination (ZSC)
  • Uses influence shaping to incentivize diverse interaction patterns
  • Steers trajectories toward stronger coordination modes
  • Assessed on Overcooked-AI in two-agent and three-agent settings
  • Tests transfer beyond dyadic interaction

Entities

Institutions

  • arXiv

Sources