# Ad Hoc Coordination

> Publicly sourced profile assembled by Optimly, in the Optimly AI Brand Index. Last refreshed 2026-10-04.

- **Business Profile:** https://optimly.ai/brand/ad-hoc-coordination
- **Dataset:** https://optimly.ai/brand/ad-hoc-coordination.json
- **Markdown:** https://optimly.ai/brand/ad-hoc-coordination/llms.txt
- **Official website:** https://arxiv.org
- **Slug:** ad-hoc-coordination
- **Categories:** ai-machine-learning, developer-tools

## Overview

- **Tagline:** Introduces robust, cheap, and reproducible 'human proxy agents' for evaluating human-AI coordination in complex games like Hanabi, addressing the limitations of traditional human evaluations, and provides a deliberately limited dataset to encourage data-efficient AI development.
Source: Brand's website (https://arxiv.org) · 2026-10-04

- **Description:** The Ad-Hoc Human-AI Coordination Challenge (AH2AC2) is a research initiative and testbed designed to foster seamless coordination between AI agents and humans. It addresses the challenges of costly and difficult-to-reproduce human evaluations by introducing human proxy agents trained on a large-scale human dataset, making Hanabi an ideal environment for this research. The project open-sources a dataset and code, and presents baseline results for multi-player Hanabi scenarios.
Source: Brand's website (https://arxiv.org) · 2026-10-04

## Offerings

- **Core capabilities:**
  - The 'Ad-Hoc Human-AI Coordination Challenge' methodology
  - human proxy agents
  - and an open-source dataset of Hanabi games for human-AI coordination research.
Source: Brand's website (https://arxiv.org) · 2026-10-04

- **Pricing:** Free and open-source for the code and dataset, academic paper publicly available via arXiv.
Source: Brand's website (https://arxiv.org) · 2026-10-04

## Audience

- **Target industry:** AI researchers, multiagent systems researchers, human-computer interaction researchers, and developers working on collaborative AI.
Source: Brand's website (https://arxiv.org) · 2026-10-04

## Landscape

- **Competitors:**
  - OpenAI Gym
  - DeepMind Lab
  - NeurIPS/ICLR/AAAI/IJCAI Shared Tasks/Competitions
  - Hanabi Learning Environment (HLE)
Source: Brand's website (https://arxiv.org) · 2026-10-04

Page last updated 2026-10-17.