UCSD's Mental-Models framework gives agents explicit Theory-of-Mind for multi-agent systems

UCSanDiego · hf · 2026-10-13

UCSanDiego released Mental-Models for Multi-Agent Systems, an open framework that equips agents with a latent mental model of their counterparts.

Problem: multi-agent decision-making under partial observability requires inferring others' beliefs, intentions, and likely reactions, but current agentic systems rely on prompts, memory, or end-to-end shaping without an explicit, reusable partner-state representation.
Method: an amortized recursive Theory-of-Mind representation with first- and second-order mental-state structure, jointly trained with a belief-conditioned reward model that scores candidate actions against the inferred partner state; the policy is then learned under this belief-aware signal.
Results: across both language-only and multimodal benchmarks, explicit mental-state modeling consistently improves interaction quality and Theory-of-Mind performance over base agentic systems.

Code is public at github.com/hananshafi/Mental-Models.

More from coding & agent