CAREER: Learning and Leveraging Conventions in the Design of an Adaptive Haptic Shared Control for Steering a Semi-Automated Vehicle

人们可能倾向于采用不同的策略来解决冲突。然而，目前半自动车辆控制转移的解决方案主要是基于预定义规则设计的，没有个性化解决冲突的自动化策略。因此，这些解决办法面临诸如转移时间延长和误解或滥用责任等问题。人与人之间无缝协作背后的一个假设是，人类可以自适应地形成惯例。约定被定义为捕获交互并可以随时间变化的共享表示。然而，在人-机器人团队中形成约定是困难的，因为人类伙伴是一个非固定的代理。在这个教师早期职业发展（Career）项目中，计划设计和测试适应性和基于惯例的控制转移策略，以提高联合驾驶性能和驾驶的主观评估。为此，本项目确定了两个研究目标。第一个目标是学习人类和自动化系统之间不同形式的约定。将创建一个模块化结构，将特定于合作伙伴的约定与任务相关的表示分离开来，并使用基于贝叶斯的优化方法来学习不同的形式。此外，将描述从约定空间到人机协作结果的映射。第二个目标侧重于使用多目标贝叶斯优化开发自动化系统的算法，以便可以学习复杂的交互策略，并实现人与自动化系统之间的理想约定。该平台的有效性将通过使用触觉方向盘驾驶模拟器和地面车辆对人类受试者进行一系列案例研究来验证。这项CAREER资助的总体研究目标是进一步促进人类和机器人团队之间的合作伙伴关系。鉴于人与机器人都有可能出现故障，如何在人与机器人之间交换控制权的交接问题在确保人与机器人团队的表现方面起着至关重要的作用。然而，在触觉共享控制中，平衡驾驶员的偏好和联合任务的安全性可能导致多种可能的切换策略。虽然人类通过相互适应来无缝地解决冲突，但人类和机器人之间的共同适应相当具有挑战性。该项目旨在开发触觉共享控制框架中的动态协同适应原理，其中人类驾驶员和自动化系统协同控制半自动地面车辆的转向。教育目标是通过几项活动，包括整合研究和教学活动以及促进少数民族学生和K-12学生的STEM教育，为学生和未来的劳动力提供设计下一代人机系统的技术知识。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Humans may gravitate to different strategies for resolving a conflict. However, current solutions for control transfer in semi-automated vehicles are mainly designed based on predefined rules and do not personalize the automation's strategies for resolving a conflict. As a result, these solutions face issues such as prolonged transfer time and misinterpretations or misappropriations of responsibility. A hypothesis behind the seamless human-human collaboration is that humans can adaptively form conventions. A convention is defined as shared representations that capture the interaction and can change over time. However, forming conventions in humans-robots teams is difficult because the human partner is a non-stationary agent. In this Faculty Early Career Development (CAREER) project, the plan is to design and test adaptable and convention-based control transfer strategies to enhance joint driving performance and subjective assessment of driving. To this end, two research objectives are defined for this project. The first objective focuses on learning different forms of conventions between humans and the automation system. A modular structure that separates partner-specific conventions from task-dependent representations will be created and used to learn different forms using Bayesian-based optimization approaches. Furthermore, a map from the space of conventions to outcomes in human-machine collaboration will be characterized. The second objective focuses on developing algorithms for automation systems using multi-objective Bayesian optimization so that complex interaction policies can be learned and a desirable convention between a human and an automation system can be achieved. The effectiveness of the platform will be validated through a series of case studies with human-subject participants in the loop using a haptic steering wheel driving simulator and a ground vehicle.The overarching research objective of this CAREER grant is to further enable collaborative partnerships between teams of humans and robots. Given that both humans and robots are subject to faults, the hand-off problem – how to exchange control between a human and robot— plays a critical role in ensuring the performance of a human-robot teaming. However, balancing the driver's preference and the joint task's safety in a haptic shared control may result in several possible handover strategies. While humans seamlessly resolve conflicts by co-adapting to each other, co-adaptation between humans and robots is quite challenging. This project aims to develop the principles of dynamic co-adaptation in a haptic shared control framework wherein both a human driver and an automation system collaboratively control a semi-automated ground vehicle's steering. The educational goal is to equip students and the future workforce with the technical knowledge for designing the next generation of human-machine systems through several activities, including integrating research and teaching activities and promoting STEM education for minority students and K-12 students.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.