CaM-Wolf: Multimodal AI Agent for Social Deduction Games
Researchers have introduced CaM-Wolf, the first AI agent for social deduction games (SDGs) like Werewolf that integrates multimodal perception and generation. Unlike previous text-based approaches, CaM-Wolf processes video inputs from other players, uses a causal-aware Reasoner trained via reinforcement learning to link observable behaviors to hidden roles, and presents itself through an animated avatar. Experiments and a user study demonstrate superior gameplay performance and enhanced human-AI interaction quality. The work is detailed in arXiv preprint 2607.26393.
Key facts
- CaM-Wolf is the first SDG agent with multimodal perception and generation.
- It processes video inputs from other players.
- Employs a causal-aware Reasoner trained via reinforcement learning.
- Presents itself through an animated avatar.
- Experiments and user study show superior gameplay and enhanced human-AI interaction.
- Published as arXiv:2607.26393.
- Social deduction games require reasoning, deception, and collaboration.
- Previous approaches were predominantly text-based.
Entities
Institutions
- arXiv