Special Issue on Autonomous Agents Modelling Other Agents

We are pleased to announce the new Special Issue on Autonomous Agents Modelling Other Agents published in Artificial Intelligence, co-edited by Stefano Albrecht, Peter Stone, and Michael Wellman. The text below was reproduced from the special issue guest editorial.

Guest Editorial

Successful interaction in multiagent systems often requires some capacity in agents to reason about the internal states and processing of other agents (e.g., their beliefs, goals, plans, or decision-making processes) and to utilise such reasoning for effective planning and decision making. Thus, the development of autonomous agents which model other agents has been a core focus in artificial intelligence research, and a recent survey by Albrecht and Stone [1] discusses the major methodologies and important open problems. As documented in the survey, this is a large area of research with a history going back several decades. However, the area is fragmented into many sub-communities with little interaction, including work on game playing, computer poker, automated negotiation, simulated robot soccer, human user modelling, human-robot interaction, commercial video games, trust and reputation, and multiagent learning. Thus, the survey sought to distil and categorise the salient approaches developed across sub-communities, and to highlight open problems shared between communities with a view to facilitating cross-fertilisation.

In the same spirit, the goal of this special issue was to provide a common forum for new technical contributions addressing open problems in autonomous agents modelling other agents. (Two previous special issues by the same editors include additional articles which consider agent modelling problems [2],[3]. See also https://mipc.inf.ed.ac.uk.) In addition, we were interested in eliciting expert perspectives about current research developments, challenges, and future directions. Authors were able to submit manuscripts until the end of December 2018. After an initial screening, 23 submitted manuscripts were put forward for review. Each submission was reviewed by three or more selected experts in a relevant area. Following two rounds of reviews, eight articles were accepted for publication in the special issue.

The first three articles fall into the category of research perspectives. Crandall [4] argues for a research agenda which places greater focus on the reverse problem of “being modelled by other agents” and the use of non-binding communication to facilitate better agent modelling, as well as a need to develop common benchmark criteria. Doshi et al. [5] provide a survey and historical account of recursive modelling methods developed in game theory and AI, and review several emerging applications as well as current limitations. Bard et al. [6] propose the Hanabi card game as a new challenge for the research community emphasising theory-of-mind reasoning, and provide an open-source Hanabi environment along with results evaluating current reinforcement learning methods.

The next five articles comprise a diverse set of technical contributions. Pereira et al. [7] propose Landmark-based approaches for goal recognition via inverse planning and show substantial speed enhancements in experiments across a large number of domains. Hayashi et al. [8] propose a sampling-based POMDP planner to estimate agent models and parameter values under partial observability, and demonstrate its efficacy in variants of a multiagent collaboration task and other tasks. Kroer and Sandholm [9] study opponents with limited lookahead in imperfect-information games, and provide game-theoretic equilibrium analyses and new algorithms for computing optimal commitment strategies under diverse opponent models. Lorini [10] considers epistemic logic reasoning in multiagent systems and proposes a new semantics based on the concept of multiagent belief base, establishing various decidability and complexity results. Singh et al. [11] propose a novel method for online human intention recognition by combining human gaze with goal recognition via inverse planning, and demonstrate improved accuracy and recognition speed in human-behavioural experiments using a multi-player board game.

Together, these articles represent a diverse collection of viewpoints and methodologies, and make significant contributions toward progress in autonomous agents modelling other agents.

The guest editors would like to thank the authors for their contributions to this special issue, and the reviewers for providing thoughtful and detailed manuscript reviews. We also thank the editor-in-chief, Patrick Doherty, and the editorial office staff at Elsevier for their assistance during the review process.

Stefano V. Albrecht
Peter Stone
Michael P. Wellman

References

  1. Stefano V. Albrecht and Peter Stone. Autonomous agents modelling other agents: A comprehensive survey and open problems. Artificial Intelligence, 258:66–95, 2018.
  2. Stefano V. Albrecht, Somchaya Liemhetcharat, and Peter Stone. Special issue on multiagent interaction without prior coordination: Guest editorial. Autonomous Agents and Multi-Agent Systems, 31:765–766, 2017.
  3. Rakesh V. Vohra and Michael P. Wellman. Foundations of multi-agent learning: Introduction to the special issue. Artificial Intelligence, 171: 363–364, 2007.
  4. Jacob W. Crandall. When autonomous agents model other agents: An appeal for altered judgment coupled with mouths, ears, and a little more tape. Artificial Intelligence, 280, 2020.
  5. Prashant Doshi, Piotr Gmytrasiewicz, and Edmund Durfee. Recursively modeling other agents for decision making: A research perspective. Artificial Intelligence, 279, 2020.
  6. Nolan Bard, Jakob N. Foerster, Sarath Chandar, Neil Burch, Marc Lanctot, H. Francis Song, Emilio Parisotto, Vincent Dumoulin, Subhodeep Moitra, Edward Hughes, Iain Dunning, Shibl Mourad, Hugo Larochelle, Marc G. Bellemare, and Michael Bowling. The Hanabi challenge: A new frontier for AI research. Artificial Intelligence, 280, 2020.
  7. Ramon Fraga Pereira, Nir Oren, and Felipe Meneguzzi. Landmark-based approaches for goal recognition as planning. Artificial Intelligence, 279, 2020.
  8. Akinobu Hayashi, Dirk Ruiken, Tadaaki Hasegawa, and Christian Goerick. Reasoning about uncertain parameters and agent behaviors through encoded experiences and belief planning. Artificial Intelligence, 280, 2020.
  9. Christian Kroer and Tuomas Sandholm. Limited lookahead in imperfect-information games. Artificial Intelligence, 283, 2020.
  10. Emiliano Lorini. Rethinking epistemic logic with belief bases. Artificial Intelligence, 282, 2020.
  11. Ronal Singh, Tim Miller, Joshua Newn, Eduardo Velloso, Frank Vetere, and Liz Sonenberg. Combining gaze and AI planning for online human intention recognition. Artificial Intelligence, 284, 2020.