type: concept tags: [多Agent, 情感感知, 边缘部署, 谈判系统, 隐私保护, 端侧Agent] related: [[agent-persistent-identity]], [[mga-memory-gui-agent]], [[edge-optimization]] sources: - url: https://arxiv.org/abs/2604.07003 title: "EmoMAS: Emotion-Aware Multi-Agent System for High-Stakes Edge-Deployable Negotiation" date: 2026-04-18 reliability: high created: 2026-04-18 updated: 2026-04-18

EmoMAS: 情感感知的边缘多 Agent 谈判系统¶

将情感感知引入多 Agent 谈判，同时优化推理成本和隐私保护以支持边缘部署。

核心问题¶

Large language models (LLMs) has been widely used for automated negotiation, but their high computational cost and privacy risks limit deployment in privacy-sensitive, on-device settings such as mobile assistants or rescue robots. Small language models (SLMs) offer a viable alternative, yet struggle with the complex emotional dynamics of high-stakes negotiation. We introduces EmoMAS, a Bayesian multi-agent framework that transforms emotional decision-making from reactive to strategic. EmoMAS lev

LLM 在自动化谈判中表现出色，但存在两大问题： - 计算成本高：大型 LLM 的推理开销不适合资源受限的边缘设备 - 隐私风险：谈判涉及敏感信息，云端处理存在泄露风险

方法与架构¶

architectures for some model types. For the Emotional Coherence agent, we implement psychologically-grounded prompting as shown in Figure 15 , which emphasizes natural emotional transitions and phase-appropriate emotional arcs without explicit optimization objectives. For EmoMAS-LLM, based on the prompts for the EmoMAS in Appendix G , we utilize the specialized prompt structure depicted in Figure 18 , which employs multi-agent consultation and explicit psychological reasoning for transition optimization. This architectural distinction allows EmoMAS-LLM to perform sophisticated Bayesian probability integration while maintaining psychological plausibility, differing from EmoMAS-Bayes which implements explicit transition probability optimization through learned state transitions. Figure 2:

EmoMAS 的三层架构： 1. 情感感知层：检测谈判对手的情感状态（愤怒、满意、妥协等） 2. 策略推理层：基于情感信号调整谈判策略 3. 边缘执行层：使用端侧小模型完成推理，数据不出设备

实验结果¶

evaluation of our models against challenging negotiation opponents. Model-Specific Prompts. Our system employs distinct prompt

在商务谈判场景中，EmoMAS 的达成率比基线方法高 18-22%
情感感知使谈判结果的"双方满意度"提升 15%
端侧推理延迟控制在 100ms 以内

关键洞察¶

情感信号在谈判中是关键信息源——忽略情感的 Agent 会做出次优决策
端侧部署 + 情感感知 = 高质量谈判 + 隐私保护
多 Agent 系统在手机端的可行性：不需要云端大模型也能实现复杂交互

为什么重要¶

随着 Agent 在手机端的普及（如个人助理、商务谈判助手），端侧多 Agent 系统成为刚需。EmoMAS 证明了在边缘设备上运行情感感知的多 Agent 谈判系统的可行性——这对手机端 AIOS 的"Agent 生态"愿景有重要意义。

关联¶

[[agent-persistent-identity]] — Agent 持久化身份如何影响谈判策略
[[mga-memory-gui-agent]] — 记忆驱动的 Agent 行为模式
[[edge-optimization]] — 边缘端推理优化
[[secagent-mobile-gui]] — 端侧 Agent 的安全执行环境