Notes深度札记
Field notes from MAC-Lab on affective AI, psychological computing, embodied emotional intelligence, student training, and research translation.MAC-Lab 关于情感智能、心理计算、具身情感智能、学生培养与科研转化的深度札记。
Field Notes for People Who Want to Understand the Route给真正想理解这条路线的人看的札记
MAC-Lab works in a space where algorithms, psychology, interaction, hardware, field validation, and public communication meet. These notes explain the lab's judgment, not just its outputs.MAC-Lab 做的是算法、心理、交互、硬件、场景验证和公共表达交汇处的问题。这些札记解释的是团队的判断方式，而不仅是成果列表。
AI psychology is not emotion labelingAI 心理计算不是给情绪贴标签
The task is not to name a feeling. The task is to understand a changing person in a changing situation.这件事的核心不是给情绪命名，而是在变化的情境中理解一个正在变化的人。
Emotion labels are useful starting points, but they are too thin for real psychological computing. A person's state is shaped by time, context, personality, language, body signals, social relationships, task pressure, and the person's willingness to be understood.情绪标签有用，但不足以支撑真实的心理计算。人的状态由时间、环境、个性、语言、生理行为信号、社会关系、任务压力以及“是否愿意被理解”共同塑造。
MAC-Lab treats Ubiquitous Psychological Computing as a full pipeline: sensing, temporal evidence, psychological profiling, risk assessment, early warning, and supportive feedback. The point is not to make the model sound certain. The point is to make uncertainty visible enough for responsible human-machine collaboration.MAC-Lab 把普适心理计算理解为一条完整链路：感知、时间证据、心理画像、风险评估、早期预警和支持性反馈。关键不是让模型显得很确定，而是让不确定性被看见，从而支持负责任的人机协同。
Overconfident diagnosis, shallow affect labels, and one-size-fits-all intervention.避免过度诊断、浅层情绪标签和一刀切干预。
Systems that combine multimodal evidence, temporal modeling, human feedback, and deployable care scenarios.建设融合多模态证据、时间建模、人类反馈和可落地关护场景的系统。
The difficult middle between a paper and a deployed system论文到系统之间，最难的是中间地带
The hard part begins after a model looks promising.模型看起来有效之后，真正困难的部分才开始。
A model can work in a benchmark and still fail in a classroom, hospital, cockpit, counseling room, or home environment. Deployment exposes sensor noise, missing data, user trust, privacy boundaries, device constraints, and maintenance cost.模型可以在基准数据上有效，却未必能在学校、医院、座舱、咨询场景或家庭环境中稳定运行。真实部署会暴露传感噪声、数据缺失、用户信任、隐私边界、设备约束和长期维护成本。
This is where MAC-Lab's experience becomes distinctive. Research translation is treated as a design discipline: define the task, the data boundary, the user's role, the evaluation protocol, the feedback loop, and the handoff mechanism before chasing model complexity.这正是 MAC-Lab 的经验价值所在。科研转化本身是一种设计能力：先定义任务、数据边界、用户角色、评测流程、反馈闭环和交付机制，再谈模型复杂度。
Keep evidence, uncertainty, and evaluation visible.让证据、不确定性和评测机制保持可见。
Make sensing, latency, devices, and data governance part of the design.把感知、延迟、设备和数据治理纳入设计。
Move beyond demos toward systems partners can trust and iterate.从演示样机走向合作方能够信任和持续迭代的系统。
A good student competition should keep growing after the final好的竞赛项目，应该在比赛之后继续生长
A medal is a checkpoint. The better question is what the project becomes next.奖项是一个节点，更重要的是项目之后还能长成什么。
Competition is a powerful entry point for undergraduate and master's students because it compresses problem definition, coding, presentation, teamwork, and scenario thinking into one cycle. But the medal is not the end of the research value.竞赛对本科生和硕士生很有价值，因为它把问题定义、代码实现、表达汇报、团队协作和场景思维压缩到一个周期里。但奖项不是科研价值的终点。
MAC-Lab tries to keep strong student work alive after the final: sharpen the research question, clean the system architecture, build reusable modules, write papers, protect intellectual property, and connect the work with a longer platform route.MAC-Lab 更看重竞赛之后的继续生长：把研究问题打磨清楚，把系统架构整理干净，把模块沉淀出来，把论文、专利、软著和长期平台路线连接起来。
Enter through projects, competitions, and real systems; learn to turn an idea into something that can be tested.从项目、竞赛和真实系统进入，学习把想法变成可验证的东西。
Turn engineering momentum into sharper problems, stronger evidence, and publishable work.把工程冲劲转化为更清楚的问题、更扎实的证据和可发表成果。
Embodied emotional intelligence is about situated response具身情感智能，核心是场景化回应能力
A good answer is not always a good response.回答正确，不等于回应合适。
The next step for AI is not only to answer correctly, but to respond appropriately in context. Robots, digital humans, smart cockpits, and companion agents must read multimodal signals, understand social timing, and express support with restraint.AI 的下一步，不只是回答正确，而是在具体情境中回应得合适。机器人、数字人、智能座舱和陪伴智能体，需要读取多模态信号，理解社会时机，并以克制、可信的方式表达支持。
Embodied Emotional Intelligence connects perception, understanding, expression, action, safety, and long-term interaction memory. It is where affective computing meets physical presence, voice, gaze, motion, and responsibility.具身情感智能连接感知、理解、表达、行动、安全和长期交互记忆。它是情感计算与身体在场、语音、视线、动作和责任边界相遇的地方。
Robots, digital humans, smart cockpits, companion agents, and mind-body health terminals.机器人、数字人、智能座舱、陪伴智能体和身心健康终端。
In care, education, mobility, and health, emotional appropriateness becomes part of system reliability.在照护、教育、交通和健康场景中，情感合适性本身就是系统可靠性的一部分。
Why a stable research spine beats short-lived topic chasing为什么长期主线比短期追题更有力量
The carriers change. The central question keeps accumulating.载体会变，但核心问题会不断积累。
MAC-Lab's route has moved from natural language processing to dialogue, affective computing, multimodal affective computing, ubiquitous psychological computing, and embodied emotional intelligence. Around 2011, Professor Xiao Sun and Professor Fuji Ren jointly came to Hefei and HFUT to advance affective computing, establishing the HFUT Affective Computing Institute with Professor Sun as director and further supporting the Anhui Key Laboratory of Affective Computing and Advanced Intelligent Machines. The carriers changed; the core question stayed steady: how can AI understand and support human state?MAC-Lab 的路线从自然语言处理、人机对话、情感计算、多模态情感计算，延伸到普适心理计算和具身情感智能。2011 年前后，孙晓教授与任福继院士共同到合肥、到合肥工业大学，联合推动情感计算方向建设，成立合肥工业大学情感计算研究所，孙晓教授担任研究所所长，并进一步参与建设情感计算与先进智能机器安徽省重点实验室、担任副主任至今。载体在变化，核心问题一直稳定：AI 如何理解并支持人的状态？
A stable spine lets students, systems, papers, patents, platforms, and partners accumulate around one expanding problem space. That accumulation is why the lab can speak to both frontier research and real industry needs without becoming either a paper factory or a pure outsourcing workshop.稳定主线让学生、系统、论文、专利、平台和合作伙伴围绕同一个不断扩展的问题空间持续积累。正因为有这种积累，实验室才能同时面向前沿科研和真实产业需求，而不是变成单一产出形态。
The route makes the lab recognizable: NLP, HCI, affective computing, AI psychology, embodied interaction, and mind-body health are not scattered keywords.这条路线让实验室有辨识度：NLP、人机交互、情感计算、AI 心理、具身交互和身心健康不是散乱关键词。
Long-term focus makes collaboration easier because partners can see where the lab has depth, systems, students, and field experience.长期聚焦让合作更容易建立信任，因为合作方能看到实验室的深度、系统、学生梯队和场景经验。
Reading Pathways推荐阅读路径
Start from training从学生培养看起
Read the competition and student-growth note, then visit People and Join.先读竞赛与学生成长札记，再看团队培养和加入我们。
Start from translation从转化逻辑看起
Read the paper-to-deployment note, then visit Projects and Contact.先读论文到系统之间的中间地带，再看科研项目和合作入口。
Start from frameworks从框架问题看起
Read the AI psychology and embodied intelligence notes, then visit Research and Frontiers.先读 AI 心理计算和具身情感智能札记，再看研究方向和前沿雷达。
Start from public value从公共价值看起
Read the stable-route note, then visit Media and the Media Kit.先读长期主线札记，再看媒体关注和传播资料。