Evaluating and modelling Hanabi-playing agents

By Joseph Walton-Rivers, P. R. Williams, R. Bartle, Diego Perez Liebana, Simon M. M. Lucas. 2017

50 citations recorded by the source corpus.

Agent modelling involves considering how other agents will behave, in order to influence your own actions. In this paper, we explore the use of agent modelling in the hidden-information, collaborative card game Hanabi. We implement a number of rule-based agents, both from the literature and of our own devising, in addition to an Information Set-Monte Carlo Tree Search (IS-MCTS) agent. We observe poor results from IS-MCTS, so construct a new, predictor version that uses a model of the agents with which it is paired. We observe a significant improvement in game-playing strength from this agent in comparison to IS-MCTS, resulting from its consideration of what the other agents in a game would do. In addition, we create a flawed rule-based agent to highlight the predictor's capabilities with such an agent.

Subfields: Artificial Intelligence, Clinical Psychology, Sociology and Political Science

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Clinical Psychology

In Sociology and Political Science

Citations in the corpus

Learned Belief Search: Efficiently Improving Policies in Partially Observable Settings (2021, background): “There has been recent work on ad-hoc team play and zeroshot coordination, in which agents are paired with unknown partners (Canaan et al., 2019; Walton-Rivers et al., 2017; Hu et al., 2020).”
The Hanabi Challenge: A New Frontier for AI Research (2019, background): “Walton-Rivers and colleagues [45] evaluated several rule-based and Monte Carlo tree search agents and observed a predictor version that modelled the other players performed better than bots without this ability.”
Bayesian Action Decoder for Deep Multi-Agent Reinforcement Learning (2018, neutral): “Walton-Rivers et al. (2017) developed a variety of Monte Carlo tree search and rule-based methods for learning to play the game, but the reported scores were roughly 50% lower than those achieved by BAD.”
Improving Policies via Search in Cooperative Partially Observable Games (2019, background): “Lastly, there is recent work on ad-hoc team play in Hanabi, in which agents get evaluated against a pool of different teammates (Canaan et al. 2019; Walton-Rivers et al. 2017).”
I See What You See: Integrating Eye Tracking into Hanabi Playing Agents (2018, background): “Walton-Rivers et al. (2017) present a comparison of several different approaches, including several based on Monte Carlo Tree Search (Browne et al. 2012), focusing on how they perform in simulated games.”
Implicit Communication of Actionable Information in Human-AI teams (2019, background): “We specifically consider the impact that Gricean conversational implicature can have on Hanabi-playing logic [52, 86].”
Evolving Agents for the Hanabi 2018 CIG Competition (2018, methodology): “ones described in [11], focusing on the probability (from its owner’s perspective) that a card is playable , or discard-able.”
Diverse Agents for Ad-Hoc Cooperation in Hanabi (2019, methodology): “[13] are the first to address the problem of playing with a diverse population of agents with different strategies in Hanabi.”
Re-determinizing Information Set Monte Carlo Tree Search in Hanabi (2019, methodology): “Eight of these are taken directly from [26], with the addition of two policies defined by the evaluation functions trained in Section VI.”