Action advising accelerates collaboration in ad hoc teams, yet existing protocols are often communication-inefficient and vulnerable to unreliable teammates. To address these limitations, we propose ACCA (Adaptive Confidence and Credibility-Aware Advising), a novel framework for robust and efficient ad hoc teamwork. ACCA integrates two key innovations: (1) a State-Importance Gating mechanism that conserves the communication budget by prioritizing advice for critical decisions, and (2) a Dynamic Credibility Model that assesses teammate reliability to select the most trustworthy guidance. Our comprehensive evaluation in a mixed cooperative-adversarial environment demonstrates that ACCA outperforms standard baselines. Ablation studies confirm its superiority, revealing a 2.6% increase in final rewards over standard advising while consuming 17% less communication budget. Critically, ACCA exhibits profound robustness against unreliable partners, outperforming the baseline by 3.3% when facing a malicious advisor and by 1.6% with a suboptimal one. This showcases its ability to maintain high performance in challenging and non-ideal collaborative settings.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Adaptive Confidence and Credibility-Aware Advising for Robust Ad Hoc Teamwork

  • Mingqian Zhu,
  • Liping Xie

摘要

Action advising accelerates collaboration in ad hoc teams, yet existing protocols are often communication-inefficient and vulnerable to unreliable teammates. To address these limitations, we propose ACCA (Adaptive Confidence and Credibility-Aware Advising), a novel framework for robust and efficient ad hoc teamwork. ACCA integrates two key innovations: (1) a State-Importance Gating mechanism that conserves the communication budget by prioritizing advice for critical decisions, and (2) a Dynamic Credibility Model that assesses teammate reliability to select the most trustworthy guidance. Our comprehensive evaluation in a mixed cooperative-adversarial environment demonstrates that ACCA outperforms standard baselines. Ablation studies confirm its superiority, revealing a 2.6% increase in final rewards over standard advising while consuming 17% less communication budget. Critically, ACCA exhibits profound robustness against unreliable partners, outperforming the baseline by 3.3% when facing a malicious advisor and by 1.6% with a suboptimal one. This showcases its ability to maintain high performance in challenging and non-ideal collaborative settings.