Reconnaissance Blind Chess

Name: Reconnaissance Blind Chess
Author: Johns Hopkins University

Johns Hopkins University

DOCUMENTED

Published in academic literature

For:Researchers & AcademicsGeneral Public & Enthusiasts

Free

Download on the App Store

App Summary

Reconnaissance Blind Chess is an app designed as an experimentation platform for artificial intelligence research, challenging users to play a variant of chess with imperfect information against AI opponents. The associated research establishes that the game models complex, real-world decision-making under uncertainty, where private sensing actions create an exponentially larger number of possible game states than in other games like poker. The authors of multiple AI tournament analyses conclude that while bot performance improves annually, developing an optimal, unexploitable strategy remains a core and unsolved research challenge.

App Screenshots

Reconnaissance Blind Chess screenshot 1 of 7

Reconnaissance Blind Chess screenshot 2 of 7

Reconnaissance Blind Chess screenshot 3 of 7

Reconnaissance Blind Chess screenshot 4 of 7

Reconnaissance Blind Chess screenshot 5 of 7

Reconnaissance Blind Chess screenshot 6 of 7

Reconnaissance Blind Chess screenshot 7 of 7

Detailed Description

Functionality & Mechanism

Developed by the Johns Hopkins University Applied Physics Laboratory, Reconnaissance Blind Chess (RBC) is an experimentation platform for artificial intelligence research. The system modifies standard chess by obscuring opponent piece locations. Each turn requires a player to first select a 3×3 board region to privately "sense," which reveals the true piece configuration within that area. The interface supports ranked and unranked games against bots, communicates captures opaquely, and provides visualization tools to manage state uncertainty.

Evidence & Research Context

The platform was designed as a standardized environment to investigate AI strategies for decision-making under conditions of extreme uncertainty and limited common knowledge.
Game-theoretic analysis demonstrates that RBC's state-space complexity is comparable to that of Go, while the average size of a player's information set is substantially larger than in games like poker.
The system serves as the official platform for the NeurIPS machine Reconnaissance Blind Chess competition, which has hosted thousands of games between dozens of unique bot submissions.
Published tournament results show progressive improvements in bot performance, yet the development of a sound, unexploitable playing algorithm remains an unsolved research challenge.

Intended Use & Scope

This system is intended for artificial intelligence researchers and computer scientists. Its primary utility is as a testbed for developing and benchmarking algorithms that operate with imperfect information. The platform implements a non-standard chess variant and is not designed for general chess instruction or recreational gameplay. Further guidance from associated publications is recommended for research applications.

Studies & Publications

3 publications

Peer-reviewed research associated with this app.

Non-Evaluative Reference

The Machine Reconnaissance Blind Chess Tournament of NeurIPS 2022

Gardner et al. (2023) · Proceedings of Machine Learning Research

Referenced in academic literature; no direct evaluation of the app

Reconnaissance Blind Chess is a game that plays like regular chess but rather than continuously observing the entire board, each player can only momentarily and privately observe selected board regions. It has imperfect information and little common knowledge. The Johns Hopkins University Applied Physics Laboratory (the game's creator) and several partners organized the third NeurIPS machine Reconnaissance Blind Chess competition in 2022 to bring people together to attempt to tackle research challenges presented by the game. 18 bots played each other in 9,180 games (60 matches per bot pair) over 4 days. The top bot exceeded the performance of all of last year's bots yet a practical, sound (unexploitable) algorithm remains unknown.

Non-Evaluative Reference

The Second NeurIPS Tournament of Reconnaissance Blind Chess

Perrotta et al. (2022) · Proceedings of Machine Learning Research

Referenced in academic literature; no direct evaluation of the app

Reconnaissance Blind Chess is an imperfect-information variant of chess with significant private information that challenges state-of-the-art algorithms. The Johns Hopkins University Applied Physics Laboratory and several organizing partners held the second NeurIPS machine Reconnaissance Blind Chess competition in 2021. 18 bots competed in 9,180 games, revealing a dominant champion with 91% wins. The top four bots in the tournament matched or exceeded the performance of the inaugural tournament's winner. However, none of the algorithms converge to an optimal, unexploitable strategy or appear to have addressed the core research challenges associated with Reconnaissance Blind Chess.