Research map / Multi-agent systems and games / Human-AI and goal-conditioned
On the Utility of Learning about Humans for Human-AI Coordination
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Train PPO with a behavior-cloned human model instead of training only with AI partners | Overcooked collaboration agents · Proxy human across the five layouts | cumulative reward | Helped |
| Train with self-play rather than a human behavior model | PPO collaboration agent · Proxy human across the five layouts | cumulative reward | Hurt |
| Train with population-based training among AI agents rather than a human behavior model | PBT collaboration agent · Proxy human across the five layouts | cumulative reward | Hurt |
| Train PPO with a behavior-cloned human model rather than AI-only partners | Overcooked collaboration agents · Real-human user study | average reward per episode | Helped |
| Train against the exact proxy-human model used at evaluation rather than the behavior-cloned proxy model | PPO collaboration agent · Proxy human across the five layouts | cumulative reward | Helped |
| Plan with the exact test partner model rather than assuming an optimal joint partner | Overcooked planner · Proxy human, two planning layouts | cumulative reward | Helped |
| Use a behavior-cloned model rather than the exact proxy model for planning | Overcooked planner · Proxy human, two planning layouts | cumulative reward | Hurt |
| Use behavior cloning directly as the agent policy instead of PPO trained with a human model | Overcooked collaboration agent · Proxy human across the five layouts | cumulative reward | Hurt |
| Use behavior cloning rather than GAIL to learn the human model | Human behavior model · Overcooked layouts | human-model performance | Helped |
| Initialize PPO training in self-play, then anneal to training with the human model (ablation) | PPO_BC · Overcooked layouts except Forced Coordination | cumulative reward | Helped |
Closest papers by meaning
- OvercookedV2: Rethinking Overcooked for Zero-Shot Coordination (ICLR 2025 · similarity 0.88)
- Learning to Cooperate with Humans using Generative Agents (NeurIPS 2024 · similarity 0.88)
- Enhancing Human Experience in Human-Agent Collaboration: A Human-Centered Modeling Approach Based on Positive Human Gain (ICLR 2024 · similarity 0.87)
- Ad-Hoc Human-AI Coordination Challenge (ICML 2025 · similarity 0.87)
- Cross-environment Cooperation Enables Zero-shot Multi-agent Coordination (ICML 2025 · similarity 0.87)
- NestRL: A Nested Training Regime for Mutual Adaptation in Human–AI Teaming (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.