Partner-Aware Hierarchical Skill Discovery for Robust Human-AI Collaboration
The article discusses a new framework called Partner-Aware Skill Discovery (PASD) designed to enhance human-AI collaboration. PASD addresses the limitations of traditional Deep Hierarchical Reinforcement Learning by focusing on partner behavior rather than solely on agent-centric rewards. The framework has been shown to improve adaptability and coordination in diverse partner scenarios, outperforming existing methods in various evaluations.
- ▪PASD learns skills conditioned on partner behavior to improve collaboration.
- ▪The framework mitigates shortcut learning by introducing a contrastive intrinsic reward.
- ▪Extensive evaluations demonstrate PASD's effectiveness in adapting to diverse partner behaviors.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.24352 (cs) [Submitted on 23 May 2026] Title:Partner-Aware Hierarchical Skill Discovery for Robust Human-AI Collaboration Authors:Adnan Ahmad, Bahareh Nakisa, Mohammad Naim Rastgoo View a PDF of the paper titled Partner-Aware Hierarchical Skill Discovery for Robust Human-AI Collaboration, by Adnan Ahmad and 1 other authors View PDF HTML (experimental) Abstract:Multi-agent collaboration, especially in human-AI teaming, requires agents that can adapt to novel partners with diverse and dynamic behaviors.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
