Collaborative Research: Human-Machine Collaboration for Design Space Exploration

协作研究：设计空间探索的人机协作

• 批准号: 1907542
• 负责人: Guy Hoffman
• 金额: $ 37.2万
• 依托单位: Cornell University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-09-01 至 2023-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1907542&HistoricalAwards=false
• 关键词: Collaborative; Research; Human; Machine; Collaboration

This project researches the collaboration between artificial intelligence (AI) agents and humans for Design Space Exploration (DSE). At the core of the research is a new perspective on designing complex systems, one in which machines complement humans instead of replacing them. The project addresses two research questions for human-machine collaborative design in the context of Design Space Exploration. First, how can engineers benefit from working with a team of separate expert AI agents, each taking a different role in the human-machine dialog? Second, how can an AI agent infer the engineer's underlying design intentions beyond explicit actions? These questions are addressed while taking into account user experience considerations and an engineer's cognitive style. The first question is approached by developing design assistants with various roles (e.g., Critic, Analyst, Historian) and different levels of initiative (proactive, reactive) and measuring their effect on design quality, diversity, learning, agent perception, and trust in the system through human-participant studies. The second question is approached by using probabilistic graphical models, including Dynamic Bayes Nets and Conditional Random Fields, taking into account explicit and implicit human behaviors, and then using Markov Decision Processes to estimate the best action. Research on user experience and the effects of cognitive style will identify the mechanisms through which these agents and benefit designers with different preferred modes of processing information.The first intellectual merit of this project is the exploration and evaluation of AI tools that significantly go beyond the state of the art in Engineering Design Space Exploration (DSE). This project will advance knowledge towards human-multi-agent DSE, towards models of probabilistic intention inference in the DSE space, and towards embodied frameworks for human-machine collaborative design. The second intellectual merit is the provision of new data sets of collaborative DSE to shed light on how designers use their explicit actions and nonverbal communication with AI assistants. The third intellectual merit is investigating how people with different cognitive styles can benefit from AI assistants in design tasks. The research questions in this project apply to a large number of design problems, and can thus have impact on many industries which engage in design, including architecture, medicine, urban planning, industrial design, and business management. Enhancing the capabilities of humans through new modes of collaboration with artificial intelligence can significantly impact how design is performed across the above areas. Additionally, the research project here provides opportunities for education and outreach during its execution. Students will be directly involved in this research agenda, and the research will be integrated with AI, robotics, user experience, and design courses.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

本项目研究人工智能(AI)智能体与人类在设计空间探索(DSE)中的合作。这项研究的核心是一种设计复杂系统的新视角，即机器补充而不是取代人类。该项目在设计空间探索的背景下解决了人机协同设计的两个研究问题。首先，工程师如何从与独立的专家人工智能代理团队合作中受益，每个代理在人机对话中扮演不同的角色？其次，人工智能代理如何在显性动作之外推断工程师潜在的设计意图？这些问题是在考虑用户体验考虑和工程师的认知风格的同时提出的。第一个问题是通过培养具有不同角色(如批评家、分析员、历史学家)和不同水平的主动性(主动、反应)的设计助理，并通过人类参与研究来衡量他们对设计质量、多样性、学习、代理感知和对系统的信任的影响。第二个问题是使用概率图模型，包括动态贝叶斯网和条件随机场，考虑了人类的显性和隐性行为，然后使用马尔可夫决策过程来估计最优动作。对用户体验和认知风格影响的研究将确定这些代理和具有不同偏好信息处理模式的设计者受益的机制。该项目的第一个智力价值是对人工智能工具的探索和评估，这远远超出了工程设计空间探索(DSE)的最新水平。该项目将把知识推向人-多智能体DSE，DSE空间中的概率意图推理模型，以及人机协同设计的具体化框架。第二个智力优势是提供了协作DSE的新数据集，以阐明设计师如何使用他们的显性动作和与AI助手的非语言交流。第三个智力优势是研究不同认知风格的人如何在设计任务中受益于人工智能助手。本项目的研究课题适用于大量的设计问题，对建筑、医药、城市规划、工业设计、企业管理等多个从事设计的行业产生影响。通过与人工智能的新协作模式增强人类的能力可以显著影响上述领域的设计执行方式。此外，这里的研究项目在执行过程中提供了教育和推广的机会。学生将直接参与这项研究议程，研究将与人工智能、机器人学、用户体验和设计课程相结合。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

Design Intention Inference for Virtual Co-Design Agents

虚拟协同设计代理的设计意图推断

• DOI: 10.1145/3383652.3423861
• 发表时间: 2020
• 期刊: IVA '20: Proceedings of the 20th ACM International Conference on Intelligent Virtual Agents
• 作者: Law, Matthew V.;Kwatra, Amritansh;Dhawan, Nikhil;Einhorn, Matthew;Rajesh, Amit;Hoffman, Guy
• 通讯作者: Hoffman, Guy

Hammers for Robots: Designing Tools for Reinforcement Learning Agents

机器人锤子：设计强化学习代理工具

• DOI: 10.1145/3461778.3462029
• 发表时间: 2021
• 期刊: DIS '21: Designing Interactive Systems Conference 2021
• 作者: Law, Matthew V;Li, Zhilong;Rajesh, Amit;Dhawan, Nikhil;Kwatra, Amritansh;Hoffman, Guy
• 通讯作者: Hoffman, Guy