Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Human-AI coordination agents
Trains agents that adapt to and collaborate with human partners, including in games like Hanabi.
125 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Human-AI and goal-conditioned, keyword cluster “communication · ai · collaboration”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
45 papers in 2023–24 and 53 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 15.2% to 13.7%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Modeling Human Ad Hoc Coordination (AAAI 2016 · 6 citations)
- 2018 · Social Norms of Cooperation With Costly Reputation Building (AAAI 2018)
- 2019 · On the Utility of Learning about Humans for Human-AI Coordination (NeurIPS 2019 · 88 citations)
- 2020 · Simplified Action Decoder for Deep Multi-Agent Reinforcement Learning (ICLR 2020 · 43 citations)
- 2021 · On the Critical Role of Conventions in Adaptive Human-AI Collaboration (ICLR 2021 · 33 citations)
- 2022 · Efficient Learning of Safe Driving Policy via Human-AI Copilot Optimization (ICLR 2022 · 15 citations)
- 2023 · Maximum Entropy Population-Based Training for Zero-Shot Human-AI Coordination (AAAI 2023 · 20 citations)
- 2024 · ProAgent: Building Proactive Cooperative Agents with Large Language Models (AAAI 2024 · 63 citations)
- 2025 · A No Free Lunch Theorem for Human-AI Collaboration (AAAI 2025 · 4 citations)
- 2026 · Collaborative Gym: A Framework for Enabling and Evaluating Human-Agent Collaboration (ICLR 2026 · 1 citation)
Earliest papers here
- Modeling Human Ad Hoc Coordination (AAAI 2016 · 6 citations)
- Social Norms of Cooperation With Costly Reputation Building (AAAI 2018)
- On the Utility of Learning about Humans for Human-AI Coordination (NeurIPS 2019 · 88 citations)
What has been tried
81 papers here report 744 measured changes (a component added, removed or swapped, compared with a baseline): helped 67%, no clear effect 9%, hurt 21%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add teammate-intention belief to analysis-based planning | Cramped Room · average reward | 184 -> 204 | Helped | ProAgent: Building Proactive Cooperative Agents with Large Language Models (AAAI 2024 · 63 citations) |
| Remove future-task anticipation | Home task completion · accuracy | Hurt | Back to the Future of Integrated Robot Systems (AAAI 2025 · 2 citations) | |
| Remove qualitative value descriptions | Desire-driven activity simulation · Rate of desire fulfillment | Hurt | Simulating Human-like Daily Activities with Desire-driven Autonomy (ICLR 2025 · 2 citations) | |
| Add morality learning that improves the LM module | Jiminy Cricket (10-game AVG) · Completion Percentage | 5.90 -> 6.75 | Helped | Human-Guided Moral Decision Making in Text-Based Games (AAAI 2024 · 2 citations) |
| Interpolate language-model representations toward the human similarity matrix | Human values · Performance across metrics | Helped | Learning Human-like Representations to Enable Learning Human Values (NeurIPS 2024 · 2 citations) | |
| Use full fine-tuning instead of decoder-only fine-tuning with human data | Multi-strategy Counter · cooperative score | 103.76 ± 1.96 -> 34.05 ± 2.01 | Hurt | Learning to Cooperate with Humans using Generative Agents (NeurIPS 2024 · 2 citations) |
| Remove adaptive policy update and perform online reinforcement learning using the inferred reward | MPE Predator-Prey · teaming performance | Hurt | Learning to Collaborate with Unknown Agents in the Absence of Reward (AAAI 2025 · 1 citation) | |
| Fine-tune Attuned on Leela moves only, excluding Maia policy moves | STT with Maia1100 junior · win-share (%) | 55.0 % ±0.5 -> 52% ± 0.5 | Hurt | Designing Skill-Compatible AI: Methodologies and Frameworks in Chess (ICLR 2024 · 1 citation) |
| Add human-led policy modification to the static interpretable policy | Optional Collaboration · maximum team performance | Helped | Designs for Enabling Collaboration in Human-Machine Teaming via Interactive and Explainable Systems (NeurIPS 2024 · 1 citation) | |
| Remove KL regularisation (λ = 0) | Hanabi self-play · self-play score | Hurt | Ad-Hoc Human-AI Coordination Challenge (ICML 2025 · 0 citations) | |
| add Q-function labeling instead of reward labeling | MetaDrive · success rate | 0.61 -> 0.82 | Helped | Robot-Gated Interactive Imitation Learning with Adaptive Intervention Mechanism (ICML 2025 · 0 citations) |
| add contrastive preference pairs instead of only behavior-cloning positive actions | MetaDrive, neural expert · success rate | 0.36 -> 0.81 | Helped | Predictive Preference Learning from Human Interventions (NeurIPS 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 38 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Baselines
- IPPO · 3 of 38 papers
- BC · 2 of 38 papers
- PVP · 2 of 38 papers
- Ensemble-DAgger · 2 of 38 papers
- Thrifty-DAgger · 2 of 38 papers
- PACE · 2 of 38 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 98 papers; from OpenAlex and the papers' affiliations).
- University of Oxford · 8 papers
- Massachusetts Institute of Technology · 7 papers
- Carnegie Mellon University · 5 papers
- Tencent · 5 papers
- Tsinghua University · 4 papers
- Georgia Institute of Technology · 4 papers
- Princeton University · 4 papers
- Harvard University · 4 papers
- Meta · 4 papers
- Google · 4 papers
Use it
Download BibTeX of 39 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Collaborative multi-agent perception (86 papers)
- Boundedly rational agent modeling (87 papers)
- Theory-of-mind social reasoning (69 papers)
- Emergent communication in referential games (50 papers)
Elsewhere in the topic
- World-model-based agents (110 papers)
- Imitation and skill learning (100 papers)
- Language-grounded reinforcement agents (96 papers)
- Embodied agent learning (93 papers)
- Intrinsically motivated exploration (86 papers)
- Interactive multi-agent trajectory prediction (75 papers)
- Learning-guided MAPF search (73 papers)
- Vision-language navigation agents (66 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).