Designing negotiation agents for international competition

• Main Author: Zhang, Huixian
• Other Authors: Bo An
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/140962
• Institution: Nanyang Technological University

Automated negotiation using autonomous agents has become an increasing popular area for Artificial Intelligence (AI) research in the recent years. The Automated Negotiating Agent Competition (ANAC) has been running since 2010 to provide a benchmark for evaluating negotiation strategies and drive the research in the automated negotiation field. The Werewolf Game League Protocol Division was introduced in 2019 where the agents play the Werewolf Game, a popular incomplete-information communication game. This paper introduces AgentCurry, an AIWolf agent developed to participate in the 2020 ANAC Werewolf Game League Protocol Division Competition. The agent follows the design similar to the model-based agent, where the agent processes the newly received game information, estimates other players role using a rule-based algorithm based on Bayesian Inference, and acts accordingly in order to build consensus with other agents, identify coalitions and cooperate with teammates to achieve victory. The testing results showed that AgentCurry outperforms the finalists in the previous competition and is good at playing the seer, the bodyguard and the werewolf role. Improvements on the playing of the possessed role are needed. Further research on role estimation model and decision-making tactics could be performed. Further testing, especially with more contest agents and a greater number of trials, is also recommended.