ToM2Vec: Continuous Theory-of-Mind Tokens for Multi-Agent Coordination

One-line takeaway

ToM2Vec enables abstract theory-of-mind reasoning in multi-agent systems, outperforming baselines across diverse environments.

ToM2Vec通过学习连续的理论思维令牌，在多智能体协作中实现对伙伴状态的抽象推理，显著提升合作与协调能力，并在多种环境下的表现优于基线模型。

Key points

ToM2Vec introduces continuous theory-of-mind tokens to enable multi-agent systems to reason about partner mental states under uncertainty. By training in a latent space with fixed reasoning tokens supervised by actual partner thoughts, it captures abstract beliefs and intentions. Experiments across diverse environments show consistent performance gains over baselines, with strong generalization to unseen partner models. Analysis confirms the tokens encode genuine theory-of-mind capabilities, not task-specific knowledge.

• ToM2Vec introduces continuous latent tokens for modeling partner mental states in multi-agent systems.
• Tokens are trained using actual partner thoughts as supervision to capture abstract beliefs and intentions.
• Outperforms baselines across collaborative, competitive, and mixed-motive environments.
• Demonstrates generalization to unseen partner models and encodes true theory-of-mind, not task-specific knowledge.

Written only from the official arXiv abstract by an open-source model, then checked automatically (every number and name must appear in the abstract). The paper is authoritative.