CAREER: Structural and Accountable Behavior Understandings and Human-centered AI Designs with Naturalistic Micromobility Riding Data

职业：结构和负责任的行为理解以及以人为中心的人工智能设计与自然微移动骑行数据

• 批准号: 2239897
• 负责人: Suining He
• 金额: $ 50万
• 依托单位: University of Connecticut
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-05-01 至 2028-04-30
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2239897&HistoricalAwards=false
• 关键词: CAREER; Structural; Accountable; Behavior; Understandings

Micromobility systems are small, lightweight vehicles that usually operate below 15 mph. Examples include regular and electric bicycles, stand-up electric scooters, and 3-wheeled scooters equipped with seats. Understanding how micromobility riders should behave to minimize conflicts with other constituents of urban traffic is essential. Using this understanding in the design of micromobility platforms and their interactions with the rider will improve safety and comfort in urban transportation and enhance public acceptance. Rider behavior is structured, consisting of maneuvers such as acceleration, turns, and dismounts and macroscopic behaviors such as the selection of locations to visit and paths to take. Building structured models of rider behavior is necessary as the existing models are difficult to use for interpreting rider accountability, and whether behaviors are acceptable to the public. To address these challenges, this project proposes to design a novel Structural and Accountable micromobility Rider Behavior Understanding System (SARBUS). SARBUS will provide the micromobility rider behavior modeling through human-centered artificial intelligence. For concrete insights and prototype development, the project will focus on the increasingly popular stand-up electric scooters (e-scooters). The model development will use data from continuous multi-modal sensors collected during real-world riding scenarios. The principal investigator will recruit and train students from under-represented groups in research, and will use the project research to promote teaching and training through student mentoring, new course development, and outreach activities.This project will be composed of two interleaved modeling thrusts. Thrust A will develop a reinforcement learning technique that interactively learns and captures the rider's macroscopic location visit and paths (through GPS logs) and microscopic maneuvering behaviors (through accelerometers and gyroscopes), as well as their structural inter-dependencies. This thrust will derive a graph representation that will be used by SARBUS to illuminate and explain the riders' decision-making process. Thrust B will build models from the motion sensor readings of the maneuver behaviors (accelerometers and gyroscopes), recorded (on-board) riding videos and the human textual annotations of those videos. These data will capture the interactions of the riders with other road users and the environment. This thrust will quantify the relationships across these modalities through graph learning, and the resulting models will reveal when rider behaviors create conflicts with other traffic elements, and identify rider accountability. The principal investigator will further conduct e-scooter rider case studies with SARBUS to understand how the structural rider behavior modeling and accountability interpretation can together benefit the riders in route selection and riding behavior with enhanced travel efficiency and fewer conflicts.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.

微移动系统是小型、轻型车辆，通常运行速度低于 15 英里/小时。例如普通自行车和电动自行车、站立式电动滑板车和配备座椅的三轮滑板车。 了解微型交通乘客应如何行事以尽量减少与城市交通其他组成部分的冲突至关重要。在微移动平台的设计及其与骑手的互动中运用这种理解将提高城市交通的安全性和舒适度，并提高公众的接受度。骑手行为是结构化的，包括加速、转弯和下车等操作以及选择要参观的地点和要采取的路径等宏观行为。 建立骑手行为的结构化模型是必要的，因为现有模型很难用来解释骑手的责任以及行为是否为公众所接受。为了应对这些挑战，该项目建议设计一种新颖的结构性和可问责的微移动乘客行为理解系统（SARBUS）。 SARBUS 将通过以人为中心的人工智能提供微移动骑手行为建模。为了获得具体的见解和原型开发，该项目将重点关注日益流行的站立式电动滑板车（e-scooters）。该模型开发将使用在真实骑行场景中收集的连续多模态传感器的数据。首席研究员将招募和培训来自代表性不足群体的学生进行研究，并将利用项目研究通过学生指导、新课程开发和推广活动来促进教学和培训。该项目将由两个交叉的建模主旨组成。 Thrust A 将开发一种强化学习技术，以交互方式学习和捕获骑手的宏观位置访问和路径（通过 GPS 日志）和微观操纵行为（通过加速计和陀螺仪），以及它们的结构相互依赖性。这种推力将得出一个图形表示，SARBUS 将使用该图形表示来阐明和解释骑手的决策过程。 Thrust B 将根据操纵行为的运动传感器读数（加速计和陀螺仪）、记录的（车载）骑行视频以及这些视频的人工文本注释来构建模型。这些数据将捕获骑手与其他道路使用者和环境的互动。这一推动力将通过图学习来量化这些模式之间的关系，生成的模型将揭示骑手行为何时与其他交通要素产生冲突，并确定骑手的责任。首席研究员将进一步与 SARBUS 进行电动滑板车骑手案例研究，以了解结构性骑手行为建模和责任解释如何共同使骑手在路线选择和骑行行为方面受益，提高出行效率并减少冲突。该奖项反映了 NSF 的法定使命通过使用基金会的智力优点和更广泛的影响审查标准进行评估，并被认为值得支持。

项目成果

Driver Maneuver Interaction Identification with Anomaly-Aware Federated Learning on Heterogeneous Feature Representations

• DOI: 10.1145/3631421
• 发表时间: 2023-12
• 期刊: Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
• 作者: Mahan Tabatabaie;Suining He

Cross-Modality Graph-based Language and Sensor Data Co-Learning of Human-Mobility Interaction

• DOI: 10.1145/3610904
• 发表时间: 2023-09
• 期刊: Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
• 作者: Mahan Tabatabaie;Suining He;Kang G. Shin

Interaction-Aware and Hierarchically-Explainable Heterogeneous Graph-based Imitation Learning for Autonomous Driving Simulation

• DOI: 10.1109/iros55552.2023.10342051
• 发表时间: 2023-10
• 期刊: 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)
• 作者: Mahan Tabatabaie;Suining He;Kang G. Shin