Design and Evaluation of a Socially Assistive Robot Schoolwork Companion for College Students with ADHD
为患有多动症的大学生设计和评估社交辅助机器人课业伙伴

Jennifer Ayissi 珍妮弗-阿伊西

University of Southern California
南加州大学

Los Angeles, USA 美国洛杉矶

socially assistive robotics, ADHD, body doubling
社交辅助机器人、多动症、身体加倍

Amy O'Connell, Ashveen Banga, Jennifer Ayissi, Nikki Yaminrafie, Ellen Ko, Andrew Le, Bailey Cislowski, and Maja Matarić. 2024. Design and Evaluation of a Socially Assistive Robot Schoolwork Companion for College Students with ADHD. In Proceedings of the 2024 ACM/IEEE International Confer ence on Human-Robot Interaction (HRI '24), March 11-14, 2024, Boulder, CO, USA. ACM, New York, NY, USA, 9 pages. https://doi.org/10.1145/3610977 3634929
Amy O'Connell, Ashveen Banga, Jennifer Ayissi, Nikki Yaminrafie, Ellen Ko, Andrew Le, Bailey Cislowski, and Maja Matarić.2024.为患有多动症的大学生设计和评估社交辅助机器人课业伙伴》（Design and Evaluation of a Socially Assistive Robot Schoolwork Companion for College Students with ADHD.In Proceedings of the 2024 ACM/IEEE International Confer ence on Human-Robot Interaction (HRI '24), March 11-14, 2024, Boulder, CO, USA.ACM, New York, NY, USA, 9 pages.https://doi.org/10.1145/3610977 3634929

This work is licensed under a Creative Commons AttributionNonCommercial-ShareAlike International 4.0 License.
本作品采用知识共享 署名-非商业性-相同方式共享 4.0 国际许可协议进行许可。
HRI '24, March 11-14, 2024, Boulder, CO, USA
HRI '24，2024 年 3 月 11-14 日，美国科罗拉多州博尔德市

(ค) 2024 Copyright held by the owner/author(s)
(ค) 2024 版权归所有者/作者所有

ACM ISBN 979-8-4007-0322-5/24/03

Figure 1: Blossom and the study system, including the tripodmounted webcam and touch screen interface
图 1：Blossom 和研究系统，包括安装在三脚架上的网络摄像头和触摸屏界面

College is challenging for all students, but especially for students with Attention Deficit Hyperactivity Disorder (ADHD). Students with ADHD experience more difficulty adjusting to college and lower performance than their peers without ADHD, evidenced by lower GPAs and graduation rates [8]. This work seeks to develop an in-dorm SAR study companion that performs idle motions to passively support college students with ADHD during schoolwork tasks. Through a three-week, within-subjects study, we evaluated schoolwork habits in a sample of college students with clinically significant symptoms of ADHD who had a socially assistive robot study companion placed in their dorm rooms. The three study conditions involved a fixed study time requirement ( 250 min ) without a robot, the same amount with a robot, and optional no minimum study time with a robot.
上大学对所有学生来说都是一项挑战，但对患有注意力缺陷多动障碍（ADHD）的学生来说尤其如此。与没有注意力缺陷多动障碍的学生相比，患有注意力缺陷多动障碍的学生在适应大学生活方面会遇到更多困难，学习成绩也会下降，这一点可以从较低的平均学分绩点（GPA）和毕业率中得到证明[8]。这项研究旨在开发一种寝室内的 SAR 学习伴侣，它能在学习任务期间做空闲动作，为患有多动症的大学生提供被动支持。通过为期三周的受试者内研究，我们评估了临床症状明显的多动症大学生的课业习惯，并在他们的寝室中放置了社交辅助机器人学习伙伴。三种研究条件包括：没有机器人时有固定的学习时间要求（250 分钟）；有机器人时有相同的学习时间要求；有机器人时没有最短学习时间要求。

of participants used the system at least once in the third condition when not required to do so, indicating that in-dorm SAR study companions are viable tools for a subset of students with ADHD. The system had an average score of 83.86 on the System Usability Scale (SUS) [5], and there was a strong correlation between participants' SUS scores and the amount of time they studied with the robot beyond the required amount.
的参与者在第三种情况下，在没有要求的情况下至少使用了一次该系统，这表明对于一部分患有多动症的学生来说，寝室内的 SAR 学习伙伴是可行的工具。该系统在系统可用性量表（SUS）[5]中的平均得分为83.86分，参与者的SUS得分与他们在规定时间之外与机器人一起学习的时间之间存在很强的相关性。

This work draws from prior research on ADHD, education, social psychology, and human-robot interaction (HRI). In this section, we introduce prior work on procrastination in ADHD and the role of co-presence and social support in facilitating action for people with and without ADHD. We connect these ideas to our work developing SAR study companions for college students with ADHD. We then explore prior technical contributions on assistive technology for students with ADHD and in-dorm robots for other populations of college students and how they relate to our work.
这项工作借鉴了之前关于多动症、教育、社会心理学和人机交互（HRI）的研究成果。在本节中，我们将介绍之前关于多动症拖延症的研究，以及共同存在和社会支持在促进多动症患者和非多动症患者行动方面的作用。我们将这些观点与我们为多动症大学生开发 SAR 学习伴侣的工作联系起来。然后，我们将探讨之前为多动症学生提供的辅助技术和为其他大学生群体提供的宿舍机器人，以及它们与我们的工作之间的关系。

Studies of undergraduate students with ADHD show a positive correlation between ADHD symptoms of inattention and general procrastination [21], mediated by executive functions (EFs) [9]. EFs are a set of cognitive processes involved in taking self-directed actions that contribute to self-regulation; they include planning, emotional and motivational regulation, goal-directed actions, and inhibition [6]. College students with ADHD self-report procrastination and related difficulties, such as delaying getting started on tasks and doing unrelated activities, as well as anxiety in response to procrastination [14]. One strategy to support students with ADHD may involve using the presence of others to accomplish tasks, referred to by the neurodivergent (ND) community as "body doubling." Despite receiving extensive media attention [25, 29], body doubling has appeared in only a few peer-reviewed publications [12, 24]. In a recent study, Eagle et al. [12] investigated how ND individuals define and use body doubling and proposed it as a way for assistive technology to support task completion/initiation for ND individuals. They collected survey responses from 220 participants ( 139 identified as having ADHD) to form a community-sourced definition of body doubling: "having someone in the room or on a call/chat ( in order to accomplish a task or be productive . The second person may be doing a different task or a similar one ( , and it is a form of accountability and helps you stay on task . The authors proposed a 2 -axis representation of body doubling; one axis represents mutuality, i.e., the body double's level of awareness, ranging from performance/accountability on one end to ambient companionship on the other, and the other axis represents space and time, ranging from no defined time/place on one end to the same time and/or place shared by the individual and body double on the other.
对患有多动症（ADHD）的本科生的研究表明，注意力不集中与一般拖延症（ADHD）症状之间存在正相关[21]，而这种正相关是由执行功能（EFs）介导的[9]。执行功能是一系列有助于自我调节的认知过程，包括计划、情绪和动机调节、目标行动和抑制[6]。患有多动症的大学生自我报告有拖延和相关的困难，如延迟开始任务和做无关的活动，以及因拖延而产生的焦虑[14]。支持多动症学生的一种策略可能是利用他人的存在来完成任务，这被神经变异者（ND）群体称为 "身体加倍"。尽管 "身体加倍 "受到了媒体的广泛关注[25, 29]，但只在少数同行评议的出版物中出现过[12, 24]。在最近的一项研究中，Eagle 等人[12] 调查了玖瑰人如何定义和使用肢体加倍，并建议将其作为一种辅助技术来支持玖瑰人完成/启动任务。他们收集了 220 名参与者（其中 139 人被确认为患有多动症）的调查反馈，形成了来自社区的身体加倍定义："为了完成任务 或提高工作效率 ，房间里有一个人 或在打电话/聊天（ 。第二个人可能在做不同的任务 或类似的任务（ ，这也是一种问责形式 ，有助于你继续完成任务 。作者提出了身体加倍的双轴表示法；一个轴表示相互性，即另一条轴代表空间和时间，一端是没有确定的时间/地点，另一端是个人和替身共享的同一时间和/或地点。

Following the recommendations of Eagle et al., we explored how socially assistive robots can act as body doubles, providing ambient companionship in a shared time and place to support college students with ADHD in schoolwork tasks.
根据 Eagle 等人的建议，我们探索了社交辅助机器人如何充当替身，在共同的时间和地点提供环境陪伴，以支持患有多动症的大学生完成学校作业。

Improved performance resulting from the presence of other individuals relates to the theory of social facilitation. This theory, first described by Triplett [33], then Allport [3], suggests that people perform better on low-complexity tasks when the task is performed alongside another person but perform worse on high-complexity tasks under the same condition (referred to as social inhibition). Zajonc later proposed the drive theory of social facilitation, stating that the mere presence of another person brings about enhanced drive and elicits dominant responses (responses with the greatest habit strength) [35]. Zajonc suggested that less effective dominant responses to complex, difficult tasks could explain why the presence of others facilitated performance on familiar and easy tasks but hindered it on novel or difficult tasks. He also suggested that the presence of an audience enhances the performance of wellpracticed responses but hinders new skill acquisition. Riether et al. [28] conducted a study of the role of social facilitation in robots and measured participant performance on simple and complex tasks in the presence of a human, an anthropomorphic robot, and alone, and found that participants performed significantly better with a robot than alone. There was no significant difference in performance between the robot and human conditions. Wechsung et al. [34] further found that, between non-anthropomorphic and very human-like robots, participants experienced decreased performance on a complex task in the presence of the very human-like robot compared to the non-anthropomorphic robot, supporting that social inhibition may be more prevalent in interactions with highly anthropomorphic robots. Participants performed best on both tasks with the non-anthropomorphic robot present, meaning that social facilitation was not observed. In this work, we explore how social facilitation and inhibition can be applied in a real-world interaction between a SAR study companion and college students with ADHD.
有其他人在场会提高学习成绩，这与社会促进理论有关。这一理论最早由特里普利特[33]提出，随后由阿波特[3]提出，认为当人们与他人一起完成低复杂度任务时，他们的表现会更好，但在相同条件下，他们在高复杂度任务中的表现会更差（被称为社交抑制）。后来，Zajonc 又提出了社会促进的驱动力理论，指出只要有另一个人在场，就会增强驱动力，引起主导反应（习惯强度最大的反应）[35]。Zajonc 认为，对复杂、困难任务的主导反应效果较差，这可以解释为什么他人的存在会促进熟悉和简单任务的表现，但却会阻碍新任务或困难任务的表现。他还认为，观众的存在会提高已练习好的反应的表现，但却会阻碍新技能的学习。Riether 等人[28] 对机器人的社会促进作用进行了研究，测量了参与者在有人类、拟人机器人和单独在场的情况下完成简单和复杂任务的表现，结果发现，参与者在有机器人在场的情况下的表现明显好于单独在场的情况。机器人和人的表现没有明显差异。Wechsung 等人[34]进一步发现，在非拟人机器人和非常拟人机器人之间，与非拟人机器人相比，在非常拟人机器人面前，参与者在复杂任务上的表现有所下降，这证明在与高度拟人化的机器人互动时，社交抑制可能更为普遍。 在非拟人机器人在场的情况下，参与者在这两项任务中的表现最好，这意味着没有观察到社会促进作用。在这项工作中，我们将探索如何在特区学习同伴与患有多动症的大学生之间的真实互动中应用社交促进和抑制作用。

Our work draws on the aforementioned theories to design a socially assistive robot study companion to assist college students with ADHD by providing companionship as they work on schoolwork tasks. The methodology is informed by the recommendations of Spiel, Williams, Hornecker, and Good [30] on user-centered and participatory design for populations with ADHD, first by involving four researchers ( ) that identify as part of the intended user population of college students with ADHD, and second by testing a low-fidelity version of the interaction among a sample of college students with ADHD symptoms to get comprehensive feedback and suggestions to inform future development of the robot.
我们的工作借鉴了上述理论，设计了一款具有社交辅助功能的机器人学习伴侣，通过在有多动症的大学生做功课时提供陪伴，为他们提供帮助。设计方法借鉴了 Spiel、Williams、Hornecker 和 Good [30] 针对多动症人群提出的以用户为中心和参与式设计的建议，首先让四名研究人员（ ）参与其中，他们都是多动症大学生目标用户群的一部分；其次，在有多动症症状的大学生样本中测试低保真版本的交互，以获得全面的反馈和建议，为机器人的未来开发提供参考。

In past work, researchers have made significant strides in exploring the potential of technology, including robots, to support students and children with ADHD. For instance, Adams et al. [2] explored the use of virtual reality technology to create a virtual classroom environment with programmed distractions, shedding light on the attentional challenges that children with ADHD face during school tasks and how technology can mediate these challenges. Fewer studies have explored the applications of SARs for students with ADHD. Berrezueta-Guzman et al. [7] created Atent@, a Robotic Assistant (RA), and a smart home environment that utilized data from two IoT devices (chair and desk) to support children with ADHD in their homework activities. Zuckerman et al. [38] designed Kip3, a social robotic device that employs a tablet-based Continuous Performance Test (CPT) to assess inattention and impulsivity in college students with ADHD. Their initial evaluation suggested that Kip3 has the potential to help students regain focus, but questions remain about its long-term effectiveness and its ability to identify
在过去的工作中，研究人员在探索包括机器人在内的技术为多动症学生和儿童提供支持的潜力方面取得了重大进展。例如，亚当斯等人[2]探索了如何利用虚拟现实技术来创建一个具有程序分心功能的虚拟教室环境，从而揭示了多动症儿童在学校学习任务中所面临的注意力挑战，以及技术如何能够缓解这些挑战。探索将 SARs 应用于多动症学生的研究较少。Berrezueta-Guzman 等人[7] 利用两个物联网设备（椅子和桌子）的数据创建了机器人助理 Atent@ 和智能家居环境，以帮助多动症儿童完成家庭作业。Zuckerman等人[38]设计了一款社交机器人设备Kip3，该设备采用基于平板电脑的连续表现测试（CPT）来评估患有多动症的大学生的注意力不集中和冲动性。他们的初步评估表明，Kip3 有可能帮助学生重新集中注意力，但对其长期有效性和识别能力仍存在疑问。
inattention in more complex, real-world situations. Our work begins to address questions of long-term effectiveness by deploying a study companion SAR to the dorms of college students with ADHD for two weeks, and a study system with no robot for a control period of one week. By allowing participants to work on their schoolwork tasks with the robot, we gained further insights into participants' needs and preferences for the robot's functionality and design.
在更复杂、更真实的情境中，注意力会更加不集中。通过在患有多动症的大学生宿舍部署为期两周的学习陪伴机器人（SAR），以及在一周的对照期内部署不含机器人的学习系统，我们的工作开始解决长期有效性的问题。通过让参与者与机器人一起完成学习任务，我们进一步了解了参与者对机器人功能和设计的需求和偏好。

Few studies have attempted to deploy robots into the dorms of college students. Abendschein, Edwards, and Edwards [1] gave robotic cats to 9 college students for six weeks, then conducted a qualitative analysis of interviews with participants to assess the lasting novelty of an in-dorm robot. They found that novelty and use of the robot companion decreased over the six-week period. We will evaluate if the same downward trend of robot use exists for study companion robots. Jeong et al. [16] deployed 35 Jibo robots to college students' dorm rooms to perform a daily positive psychology intervention with the participant. They found that after completing the study, participants' psychological well-being, mood, and readiness to change behavior improved significantly. Our study followed similar recruitment and deployment methods but with duration of use and perceived usefulness as the primary measures of success.
很少有研究尝试在大学生宿舍中部署机器人。Abendschein、Edwards和Edwards[1]为9名大学生提供了为期六周的机器猫，然后对参与者的访谈进行了定性分析，以评估宿舍机器人的持久新鲜感。他们发现，在六周的时间里，机器人伴侣的新奇感和使用率都有所下降。我们将评估学习陪伴机器人的使用是否也存在同样的下降趋势。Jeong 等人[16]在大学生寝室中安装了 35 个 Jibo 机器人，每天与被试者一起进行积极心理学干预。他们发现，在完成研究后，参与者的心理健康、情绪和改变行为的意愿都得到了显著改善。我们的研究采用了类似的招募和部署方法，但将使用时间和感知有用性作为衡量成功与否的主要标准。

Consistent with the exploratory nature of this study, we sought to answer the following research questions:
根据本研究的探索性质，我们试图回答以下研究问题：

RQ1: Will college students with ADHD find a SAR useful as a study companion, as measured by quantitative surveys and voluntary use?
问题 1：根据定量调查和自愿使用情况，患有多动症的大学生是否会认为 SAR 是有用的学习伴侣？

RQ2: What features of the system will students with ADHD like and dislike during study sessions? How would students like the robot to behave, and how common are those preferences among students?
问题 2：多动症学生在学习过程中喜欢和不喜欢系统的哪些功能？学生希望机器人如何表现？

RQ3: What features could be added to make the robot study companion more useful to students with ADHD?
问题 3：可以增加哪些功能，使机器人学习伴侣对多动症学生更有用？

We aimed to involve end users in the design process early while simultaneously creating an initial design sophisticated enough to give users an idea of how the study companion might look and function. Therefore, we chose to use an existing robot embodiment that could be easily adapted for our deployment. Specifically, we used the Blossom open-source 3D-printed robotics research platform developed by Suguitan and Hoffman [32], which is inexpensive and could be quickly fabricated at the scale needed for this deployment We chose a grey crocheted exterior with button features, similar to the crocheted exterior described in [32], to give Blossom a simple and engaging appearance that is not too distracting. We used a basic version of the Blossom robot in this study to create a minimum working design for initial user testing and obtaining user feedback on the robot's embodiment to inform future iterations of the physical robot design.
我们的目标是让最终用户尽早参与到设计过程中来，同时设计出足够复杂的初始设计，让用户了解研究伴侣的外观和功能。因此，我们选择使用现有的机器人，这样可以很容易地适应我们的部署。具体来说，我们使用了由 Suguitan 和 Hoffman 开发的 Blossom 开源 3D 打印机器人研究平台[32]，该平台价格低廉，可以快速制造出部署所需的规模。我们选择了灰色钩针编织的外观，带有按钮功能，类似于[32]中描述的钩针编织外观，使 Blossom 的外观简洁迷人，不会太分散注意力。在这项研究中，我们使用了一个基本版本的 Blossom 机器人，为初步的用户测试创建了一个最低限度的工作设计，并获得了用户对机器人体现的反馈，为未来的实体机器人设计迭代提供了参考。
The entire system, pictured in Figures 1 and 2, consisted of a Blossom robot, a tabletop tripod and webcam, and a Raspberry Pi 4 computer connected to a 7 -inch touch screen display. A simple user interface (UI) displayed on the touch screen allowed the participant to start, pause, continue, and end schoolwork sessions. They were also able to preview the webcam input before starting a session to confirm that they were visible in the video frame. The system recorded a log of UI button press events throughout the deployment.
整个系统（如图 1 和图 2 所示）包括一个 Blossom 机器人、一个桌面三脚架和网络摄像头，以及一台连接到 7 英寸触摸显示屏的 Raspberry Pi 4 电脑。触摸屏上显示的简单用户界面（UI）允许参与者开始、暂停、继续和结束学校作业。他们还能在开始会话前预览网络摄像头输入，以确认自己在视频帧中是否可见。在整个部署过程中，系统记录了用户界面按键事件日志。

Figure 2: The Blossom robot, Logitech webcam, and 7-inch touch screen UI connected to a Raspberry Pi 4 computer
图 2：与 Raspberry Pi 4 电脑相连的 Blossom 机器人、罗技网络摄像头和 7 英寸触摸屏用户界面

During an active schoolwork session, the touch screen displays a 25 -minute timer that counts down to 0:00. When 25 minutes have elapsed, or if the user elects to end the session early, the video and audio recorded from the webcam are stored in AWS S3 cloud storage. The 25 -minute duration was chosen because prior research has shown that students with ADHD commonly use the Pomodoro technique [10], which involves working in 25 -minute sessions followed by 5 -minute breaks to ensure that focus-intensive tasks are interspersed with short breaks . This method is consistent with the "time on-time off" approach to allocating schoolwork time that many practitioners recommend for students with ADHD [23]. To give users a basic idea of how the robot might behave during a study session, we created a simple behavior policy that involved executing one of three types of hard-coded motions at random intervals throughout the study session. While a web camera recorded video and audio of the sessions for post-study analysis, the participant's actions did not influence Blossom's behavior. This was communicated to participants at their system setup appointment. During sessions, they were permitted to work on any schoolwork task, including any assignments or study activities. They were not permitted to engage in activities that were not directly related to their classes, such as leisure or extracurricular activities.
在活跃的学习过程中，触摸屏会显示一个 25 分钟的计时器，倒计时到 0:00。当 25 分钟过去后，或者如果用户选择提前结束会话，网络摄像头录制的视频和音频将存储在 AWS S3 云存储中。之所以选择 25 分钟的持续时间，是因为先前的研究表明，患有多动症的学生通常使用 Pomodoro 技术[10]，即在 25 分钟的会话中工作，然后休息 5 分钟，以确保在注意力密集型任务中穿插短暂的休息 。这种方法与许多从业者推荐给多动症学生的 "上机-下机 "课业时间分配方法一致[23]。为了让用户对机器人在学习过程中的行为有一个基本概念，我们创建了一个简单的行为策略，即在整个学习过程中以随机间隔执行三种硬编码动作中的一种。虽然网络摄像头记录了学习过程中的视频和音频，供学习后分析之用，但参与者的行为并不会影响 Blossom 的行为。这一点已在预约系统设置时告知参与者。在学习过程中，他们可以处理任何学校作业，包括任何作业或学习活动。他们不得从事与课程无直接关系的活动，如休闲或课外活动。

For the duration of each study session, the robot performed a set of idle motions to maintain a lifelike and friendly presence. Studies have found that idle motions, small, lifelike movements that robots and agents perform during periods of inactivity [11], can help robots appear more friendly [4] and entertaining [19]. Following the idle motion designs described in [11], three types of idle motions were implemented for this study: gaze shift, posture sway, and sigh. We
在每次研究过程中，机器人都会做一系列空闲动作，以保持栩栩如生的友好形象。研究发现，空闲动作是机器人和代理在不活动期间所做的细小而逼真的动作[11]，可以帮助机器人显得更加友好[4]和更具娱乐性[19]。根据文献[11]中描述的空闲动作设计，本研究实施了三种类型的空闲动作：目光移动、姿势摇摆和叹息。我们
chose to follow the idle motions described by [11] because they were clearly defined at the actuator level, making them easy to replicate on Blossom, and had been validated as portraying low social verification for a robot in a similar task companion context. Sighs were implemented by actuating the head to its maximum height over a 2 -second period then lowering the head back to a neutral height over another 2 -second period. The sighing motion repeated every 60 seconds for the duration of the interaction. Idle gaze shifts were implemented by actuating the head pitch and whole body yaw to one of three predefined values (pitch: chin down, neutral, chin up, yaw: turn left, look straight ahead, turn right) over a 0.5 -second period. Idle gaze shifts were executed at random intervals that varied between 15 and 22 seconds. The posture sway motions were implemented by actuating the head roll to one of three predefined positions (tilt left, tilt right, neutral) over a 1-second period. Posture sway motions were executed at random intervals that ranged between 20 and 30 seconds. All idle motions were implemented as actuations of each of Blossom's four motors to hardcoded goal positions. Two of the authors with ADHD completed test sessions with the robot to fine-tune the speed and exaggeration of the idle motions to avoid motions that would be too invasive or distracting during a study session.
我们选择沿用文献[11]中描述的空闲动作，因为这些动作在执行器层面上定义明确，易于在 Blossom 上复制，而且经过验证，在类似的任务陪伴环境中，机器人的社交验证程度较低。叹气是通过在 2 秒钟内将头部推至最大高度，然后再在 2 秒钟内将头部降低至中立高度来实现的。在整个互动过程中，叹气动作每 60 秒重复一次。在 0.5 秒的时间内，通过将头部俯仰和整个身体偏转至三个预定义值之一（俯仰：下巴向下、中立、下巴向上；偏转：左转、直视前方、右转）来实现空闲注视转换。空闲时的视线转移以 15 到 22 秒之间的随机间隔进行。姿势摇摆动作是通过在 1 秒钟内将头部滚动到三个预定位置（左倾、右倾、中立）之一来实现的。姿态摇摆动作以 20 至 30 秒的随机间隔执行。所有空闲动作都是通过 Blossom 的四个电机分别执行到硬编码的目标位置。两位患有多动症的作者与机器人一起完成了测试，对空闲动作的速度和夸张程度进行了微调，以避免在研究过程中出现过于侵入性或分散注意力的动作。

To evaluate the study companion robot in a dorm environment, we completed a user study, deploying study systems to 11 college students with clinically significant ADHD symptoms.
为了评估学习陪伴机器人在宿舍环境中的使用情况，我们完成了一项用户研究，为 11 名有明显临床多动症症状的大学生配置了学习系统。

3.4.1 Study Design. To evaluate the study companion robot's performance in long-term, in-dorm conditions, we conducted a threeweek within-subjects user study, in which each week corresponded to one of the following three conditions:
3.4.1 研究设计为了评估学习陪伴机器人在长期宿舍条件下的表现，我们进行了一项为期三周的主体内用户研究，每周对应以下三种条件之一：

Condition A: Participants were asked to complete a minimum of 250 minutes of schoolwork ( 10 full sessions) with the Raspberry Pi , touch screen interface, and webcam, but no robot. The touch screen interface was identical to conditions B and C.
条件 A：要求参与者使用 Raspberry Pi、触摸屏界面和网络摄像头完成至少 250 分钟的学校作业（10 节完整课程），但不使用机器人。触摸屏界面与条件 B 和 C 相同。

Condition B: Participants were instructed to complete a minimum of 50 minutes of schoolwork (equal to 2 full sessions) per day with the robot for 5 days during the week.
条件 B：要求参与者在一周内的 5 天里，每天与机器人一起完成至少 50 分钟的学校作业（相当于两节完整的课程）。

Condition C: Participants were given no minimum number of sessions to complete and permitted to leave the cover over the webcam during schoolwork sessions with the robot.
条件 C：参与者没有完成课程的最低次数限制，允许他们在与机器人一起完成学校作业的过程中不遮盖网络摄像头。

To control for ordering effects among the three study conditions, the participants were separated into two groups that determined the order in which they proceeded through the study conditions. Six participants were assigned to condition A in the first week, condition B in the second week, and condition C in the third week. The other five participants were assigned to condition B in the first week, condition C in the second week, and condition in the third week of the study. Participants began the study on different days across one week. The start and end of each week was determined based on the date that the participant began the study. Condition C always followed condition B to avoid re-introducing novelty and learning effects in week C. The study was structured to encourage participants to practice using the robot daily in week B, so we could examine if daily use continued voluntarily in week C. Participants were asked to complete at least 250 minutes of schoolwork with and without the robot to collect an adequate amount of session data for post-study analysis of video and audio features.
为了控制三种学习条件之间的排序效应，参与者被分为两组，由两组决定他们在学习条件中的顺序。六名参与者在第一周被分配到条件 A，第二周被分配到条件 B，第三周被分配到条件 C。另外五名参与者在第一周被分配到条件 B，第二周被分配到条件 C，第三周被分配到条件 。参与者在一周内的不同日子开始研究。每周的开始和结束时间根据参与者开始研究的日期确定。为了避免在第C周再次引入新奇感和学习效应，我们将条件C安排在条件B之后。研究的目的是鼓励参与者在第B周每天练习使用机器人，这样我们就可以在第C周研究参与者是否自愿继续每天使用机器人。我们要求参与者在使用和不使用机器人的情况下完成至少250分钟的学校作业，以便收集足够的会话数据，用于研究后的视频和音频特征分析。

3.4.2 Participants. We recruited participants for the study by sending out an initial online screening survey for interested students through university mailing lists. In the survey, candidates answered a set of questions and completed the Adult ADHD Self Report Scale (ASRS) [17] to determine their eligibility to participate.
3.4.2 参与者。我们通过大学邮件列表向有兴趣的学生发送了一份初步的在线筛选调查，以此招募研究参与者。在调查中，候选人回答了一系列问题，并填写了成人多动症自我报告量表（ASRS）[17]，以确定其参与资格。

Participants were selected according to the following inclusion criteria: currently enrolled as a university student, 18 years of age or older, normal or corrected normal vision and hearing, proficient in English, have a private workspace in their residence where they primarily complete most of their schoolwork, and have a score of at least 4 on the ASRS. Before taking the survey, candidates read and signed a statement of consent to collect their information for screening. Candidates that met the study criteria were emailed in the order in which their responses were received. They received a brief description of the study procedure, and those who confirmed their interest were sent a link to schedule an initial setup appointment. Due to the narrow inclusion criteria and timeintensive study procedure, we were able to recruit only a small sample of twelve participants for the study. We opted to proceed with this small sample rather than relax the inclusion criteria and recruit members that do not represent the intended population of college students with ADHD, following Spiel et al.'s [30] recruiting recommendations for populations with ADHD. A sample size of twelve is also comparable to that of other long-term in-dorm HRI studies . One participant (P11) dropped out in the final week of the study; eleven participants completed the study. Participants that completed the study identified as: 6 Female, 5 Male; 7 Asian, 1 African-American, 2 Hispanic Latino, 1 did not disclose; the ages ranged from 18 to . Their current level of education being pursued was: 7 Bachelor's (undergraduate), 4 Master's (graduate). Participants' majors were: 2 Psychology, 1 Communication, 1 Neuroscience, 1 Cognitive Science, 1 Biochemistry, 1 International Relations, 1 Computer Science, 1 Human Biology, 1 Health Promotion and Disease Prevention, 1 Health and Human Sciences, 1 Machine Learning, 1 Electrical Engineering. 5 participants reported having previously used a time tracking productivity app to complete schoolwork. Participants received a digital Amazon gift card upon completing the study.
参与者的筛选标准如下：在校大学生、18 岁或以上、视力和听力正常或矫正正常、精通英语、住所内有私人工作空间（主要在此完成大部分学业）、ASRS 得分至少为 4 分。在参加调查之前，候选人阅读并签署了一份同意收集其信息以进行筛选的声明。符合研究标准的候选人会按照收到回复的顺序收到电子邮件。他们收到了关于研究程序的简要说明，确认有兴趣的候选人还收到了一个链接，以便安排初次设置预约。由于纳入标准较窄，研究程序耗时较长，我们只能招募到 12 名参与者作为研究的小样本。根据斯皮尔等人[30]针对多动症人群的招募建议，我们选择使用这个小样本，而不是放宽纳入标准，招募那些不能代表患有多动症的大学生群体的成员。12人的样本量也与其他长期寝室HRI研究 的样本量相当。一名参与者（P11）在研究的最后一周退出，11名参与者完成了研究。完成研究的参与者身份如下6 名女性，5 名男性；7 名亚洲人，1 名非洲裔美国人，2 名拉美裔西班牙人，1 名未透露身份；年龄在 18 岁到 之间。他们目前所受的教育程度为：7 名本科生，4 名硕士生。 参与者的专业是心理学 2 名、传播学 1 名、神经科学 1 名、认知科学 1 名、生物化学 1 名、国际关系 1 名、计算机科学 1 名、人类生物学 1 名、健康促进与疾病预防 1 名、健康与人文科学 1 名、机器学习 1 名、电子工程 1 名。5名参与者表示以前曾使用过时间追踪应用程序来完成学校作业。完成研究后，参与者将获得一张 亚马逊数字礼品卡。

3.4.3 Procedure. The study was approved by the University Institutional Review Board (IRB #UP-22-01073). Participants who met the inclusion criteria were invited to schedule a setup appointment where a researcher traveled to the student's residence and set up the system. The participants reviewed a consent form, consented to participate in the study, then completed a pre-study survey. After setting up the system, the researcher showed the participant how to use the system then explained the study procedure.
3.4.3 程序。本研究获得了大学机构审查委员会（IRB #UP-22-01073）的批准。符合纳入标准的参与者被邀请预约设置时间，由研究人员前往学生住所设置系统。参与者阅读同意书，同意参与研究，然后填写一份研究前调查表。设置完系统后，研究人员向参与者演示了如何使用系统，然后解释了研究程序。

All participants were asked to complete a short post-session questionnaire after each study session completed with the system. Between conditions A and , a researcher returned to the participant's dorm to either set up a robot and connect it to the study system or collect the robot, depending on the starting condition. The study setup remained unchanged in weeks B and C (study system + robot); seven days after starting condition B, participants
所有参与者在使用系统完成每个学习环节后，都要填写一份简短的会后调查问卷。在条件 A 和 之间，研究人员会根据起始条件，返回参与者宿舍安装机器人并将其连接到研究系统，或者收集机器人。B 周和 C 周的研究设置保持不变（研究系统+ 机器人）；在条件 B 开始七天后，参与者
received an email notifying them to begin week C and a link to complete a mid-study survey.
他们收到了一封电子邮件，通知他们开始第 C 周的学习，并提供了完成中期调查的链接。

Upon completing the third week of the study, participants were given the option to have a researcher travel to their dorm to collect the system or to break down the system themselves and return it to the research lab. In both cases, during the final appointment, participants completed a post-study questionnaire and a semi-structured interview about their experience.
在完成第三周的研究后，参与者可以选择让研究人员前往其宿舍收集系统，或者自己拆卸系统并将其送回研究实验室。在这两种情况下，参与者都要在最后一次约会期间填写一份研究后调查问卷，并接受一次有关其经历的半结构化访谈。

3.4.4 Measures. In the pre-study survey, participants answered questions about their demographic information, degree program and area of study, the amount of credits they were currently enrolled in, and their expected graduation year. They also completed the Executive Skills Questionnaire-Revised (ESQ-R) [31], and the Negative Attitude towards Robots Scale (NARS) [22]. In the postsession questionnaire, participants wrote a description of the task they worked on during the session and filled out the NASA Task Load Index (TLX) [15] for that task. The TLX questionnaire is a selfreport scale that estimates the cognitive load based on participant ratings. On the mid-study survey, participants repeated the NARS questionnaire.
3.4.4 测量方法在研究前调查中，参与者回答了有关他们的人口信息、学位课程和学习领域、目前修读的学分以及预计毕业年份等问题。他们还填写了执行技能问卷-修订版（ESQ-R）[31]和机器人负面态度量表（NARS）[22]。在课后调查问卷中，参与者写下了他们在课上所做任务的描述，并填写了该任务的美国宇航局任务负荷指数（TLX）[15]。TLX 问卷是一个自我报告量表，根据参与者的评分估算认知负荷。在研究中期调查中，参与者重复了 NARS 问卷。

In the post-study survey, participants completed the NARS and ESQ-R questionnaires. Additionally, the post-study survey included some background questions about the participant's prior experiences, such as their experience with productivity apps. Finally, a researcher conducted semi-structured interviews with each participant to gather more information about their experience with the robot. These interviews were conducted by one researcher who followed a script of questions, asking unscripted follow-up questions based on the participant's responses. Participants were asked what they liked and disliked about Blossom and what they thought of Blossom's behavior during the study sessions. The researcher then asked about any changes the participant would make to the robot to improve it and what features they think would be useful in an in-dorm robot. During this ideating stage, participants were encouraged not to worry about whether their ideas were practical or technologically feasible.
在学习后调查中，参与者填写了 NARS 和 ESQ-R 问卷。此外，研究后调查还包括一些有关参与者先前经历的背景问题，例如他们使用生产力应用程序的经历。最后，一名研究人员对每位参与者进行了半结构化访谈，以收集更多有关他们使用机器人体验的信息。这些访谈由一名研究人员进行，他按照问题脚本进行提问，并根据参与者的回答提出没有脚本的后续问题。参与者被问及他们喜欢和不喜欢 Blossom 的哪些方面，以及他们对 Blossom 在学习过程中的行为有何看法。然后，研究人员会询问参与者会对机器人做出哪些改进，以及他们认为宿舍机器人应该具备哪些功能。在这一构思阶段，研究人员鼓励参与者不要担心他们的想法是否实用或在技术上是否可行。

3.4.5 Analysis. We used the log of UI events from each system to extract information about the study sessions, such as the total amount of time each participant spent in active sessions with the system under each condition and when the sessions took place.
3.4.5 分析。我们使用每个系统的用户界面事件日志来提取有关研究会话的信息，例如每个参与者在每个条件下与系统进行活动会话的总时间，以及会话发生的时间。

We calculated the SUS scores for each participant from the poststudy survey. We performed paired Wilcoxon signed-rank tests on participants' pre-study and post-study ESQ-R scores. We performed a Wilcoxon signed-rank test between the NASA-TLX scores of all participants during condition A and the combined NASA-TLX scores of all participants during conditions B and C . We performed a repeated measures ANOVA test between participants' pre-, mid-, and post-experiment NARS scores.
我们计算了每位学员在学习后调查中的 SUS 分数。我们对学员在学习前和学习后的 ESQ-R 分数进行了配对 Wilcoxon 符号秩检验。我们对条件 A 中所有参与者的 NASA-TLX 分数与条件 B 和条件 C 中所有参与者的 NASA-TLX 分数进行了 Wilcoxon 符号秩检验。我们对参与者在实验前、实验中和实验后的 NARS 分数进行了重复测量方差分析。

Audio recordings of the post-study interviews were transcribed with OpenAI's Whisper speech recognition model [26], then verified by one author for correctness and separated into individual sentences. To eliminate the potential of introducing bias, two coders with no prior involvement in the project separately reviewed each transcript and identified emerging themes. The coders then met to discuss their individual findings and formed a unified list of mutual themes. They reviewed the transcripts and coded each sentence according to each theme on the unified list to identify the prevalence of each theme across all participants. Codes were assigned with inter-rater reliability (mean Cohen's ). Finally, researchers identified the themes related to each research question.
研究后访谈的录音使用 OpenAI 的 Whisper 语音识别模型[26]进行转录，然后由一位作者核实其正确性，并将其分成单个句子。为了消除引入偏见的可能性，两名事先未参与该项目的编码员分别审阅了每份记录誊本，并确定了新出现的主题。然后，两位编码员开会讨论各自的发现，并形成一份统一的共同主题清单。他们审阅了记录誊本，并根据统一列表中的每个主题对每个句子进行编码，以确定每个主题在所有参与者中的普遍程度。编码的分配具有评分者之间的可靠性（平均 Cohen's ）。最后，研究人员确定了与每个研究问题相关的主题。

The participants gave the system an average SUS score of 83.864, earning an "A" in usability ( 80.3 or higher). Spearman's rank correlation was computed to assess the relationship between SUS score and total time the participant used the study system under condition C. There was a positive correlation between the two measures , meaning perceived system usability was a reliable predictor of continued use.
参与者给系统打出的平均 SUS 分数为 83.864，在可用性方面获得了 "A"（80.3 或更高）。我们计算了斯皮尔曼等级相关性，以评估在条件 C 下 SUS 分数与受试者使用研究系统的总时间之间的关系。这两个测量值之间存在着 的正相关关系，这意味着感知到的系统可用性是持续使用的可靠预测因素。

Of 11 college students who participated in the study, elected to have schoolwork sessions with the SAR companion system under condition when not required to do so. Participants spent an average of 93.041 minutes ( ) in active sessions with the study companion robot under condition C. Figure 3 shows the number of users that started active sessions each day in the voluntary condition. Figure 4 shows the distribution of session start times in each of the three conditions.
在参与研究的 11 名大学生中， 选择在条件 下与 SAR 伴学系统进行课业辅导，而没有要求这样做。图 3 显示了在自愿条件下每天开始主动会话的用户数量。图 4 显示了三种情况下开始活动时间的分布。

Results of the Wilcoxon signed-rank tests showed no significant increase between participants' pre-study and post-study ESQR scores ( ), indicating that the minimal robot design had no measurable effect on the participant's executive functions.
Wilcoxon 符号秩检验的结果显示，参与者在研究前和研究后的 ESQR 分数没有显著增加（ ），这表明最小机器人设计对参与者的执行功能没有产生可测量的影响。

There was no significant difference between each participant's average post-session NASA-TLX scores during week A (no robot present) and during weeks and combined (robot present) ( . There was also no significant change in NARS scores pre-, mid-, and post-experiment ( ).
在 A 周（无机器人在场）和 和 两周（有机器人在场）期间，每位参与者的会后 NASA-TLX 平均得分没有明显差异（ ）。NARS 分数在实验前、实验中和实验后也没有明显变化 ( )。

To confirm that the order of the study conditions did not affect participant outcomes, we computed the difference between preand post-study scores on the NARS and ESQR for each participant and compared outcomes between the two ordering groups (group 1: cond. ) (group 2: cond. ). We performed Mann-Whitney Tests and found no significant difference between groups 1 and 2 (ESQR: ; NARS: , indicating that ordering effects were not present.
为了确认研究条件的顺序不会影响参与者的结果，我们计算了每位参与者在 NARS 和 ESQR 上研究前后得分的差异，并比较了两个排序组（第 1 组：cond. ）（第 2 组：cond. ）的结果。我们进行了 Mann-Whitney 测试，发现第 1 组和第 2 组之间没有显著差异（ESQR： ；NARS： ）： ，表明不存在排序效应。

In this section, we report the themes that emerged from the poststudy interviews and how they address RQ2 and RQ3.
在本节中，我们将报告研究后访谈中出现的主题，以及这些主题是如何解决问题 2 和问题 3 的。

4.2.1 Studying Behavior and Attitude Toward Studying. Participants listed a wide variety of ways that Blossom affected their studying. Table 1 shows the most common ideas that participants expressed. We broadly categorize the most common themes in this area as related to the participant's ability to focus, motivation to study, and ability to manage their time.
4.2.1 学习行为和学习态度。学员们列举了 Blossom 对他们学习的各种影响。表 1 列出了学员们最常表达的观点。我们将这一领域最常见的主题大致归类为与学员的专注能力、学习动力和时间管理能力有关。

Participants suggested that a sense of companionship with the robot made it easier for them to study. For example, P4 compared studying with Blossom to being in a library: "it's kind of like when
学员们表示，与机器人为伴的感觉让他们更容易学习。例如，P4 将与 Blossom 一起学习比作在图书馆："这有点像

Figure 3: Daily use of the in-dorm study companion robot under condition (voluntary use)
图 3：在 （自愿使用）条件下宿舍内学习陪伴机器人的日常使用情况

Figure 4: Session start times, by hour, in each condition
图 4：每种情况下按小时分列的会议开始时间

you're working at a library and you see everyone working around you, doing little movements, having their iPads out." Concerning staying focused during study sessions, participants stated that Blossom's constant movement kept them engaged and that trying not to pay attention to the robot's movements made them pay more attention to their work. P5 stated "It was making some weird movements, but that has inspired me, it has motivated me to concentrate more on my studies and not get distracted by the movements. So that has, you know, that has improved my willpower that I should not concentrate on the robot." Participants also thought it was easier to focus when they could verbalize their thoughts to the robot. P7 stated "I talk a lot to myself while I do my work, so it was kind of fun to just talk, even though I know Blossom couldn't talk back. It made me feel like I was able to be more on task. I felt like my thoughts came more fluidly, and I felt more comfortable because I guess it wasn't a person there, but [the robot] made me feel like I wasn't completely talking to a wall. I just kind of felt like there's another presence, I guess."
在图书馆工作时，你会看到周围的人都在工作，做着小动作，拿出他们的 iPad"。关于在学习过程中保持注意力集中的问题，学员们表示，Blossom 的持续运动让他们全神贯注，尽量不去注意机器人的动作让他们更加专注于自己的工作。P5 说："它在做一些奇怪的动作，但这给了我启发，促使我更加专注于学习，不被机器人的动作分心。所以，你知道，这提高了我的意志力，我不应该把注意力放在机器人身上"。学员们还认为，当他们可以向机器人说出自己的想法时，注意力会更容易集中。P7 说："我在工作时经常自言自语，所以尽管我知道 Blossom 不会回话，但我还是觉得说话很有趣。这让我感觉自己能更专注于工作。我觉得我的想法更流畅了，我感觉更舒服了，因为我想这不是一个人在那里，但（机器人）让我觉得我不是完全在对着一堵墙说话。我想，我只是觉得有另一种存在"。

Blossom also positively impacted the participants' self-reported motivation to study. They expressed that interacting with Blossom made the schoolwork more enjoyable and that Blossom's subtle companionship and presence encouraged them to spend more time on schoolwork. When asked if it was easier to start working with the robot compared to the system with no robot, P3 explained that "I felt like it was slightly easier because it was more like a fun activity, turning on the system and having that robot next to me doing its little thing while studying. So I would say it was definitely more enjoyable
Blossom 还对参与者自我报告的学习动机产生了积极影响。他们表示，与 Blossom 的互动让学习变得更加愉快，Blossom 微妙的陪伴和存在鼓励他们花更多时间在学习上。当被问及与没有机器人的系统相比，开始与机器人一起工作是否更容易时，P3 解释说："我觉得稍微容易一些，因为这更像是一种有趣的活动，打开系统，让机器人在我旁边一边学习一边做它的小动作。因此，我认为这无疑更令人愉快
Table 1: Participant responses related to their behavior during study sessions and overall attitude toward studying
表 1：学员对学习期间的行为和总体学习态度的回答

[than the system with no robot] and I made sure that I was getting a few hours of work done every day."
[与没有机器人的系统相比]，我确保每天都能完成几个小时的工作"。

Finally, participants suggested that the study companion led them to manage their study time effectively. Blossom reminded them to use time wisely, which led them to take fewer unnecessary breaks. P1 stated "I started noticing I was like, 'I know I'm gonna get a lot of work done in the time that [Blossom]'s on.' So I'd be like, 'Okay, cool. I'm gonna sit down, I'm gonna get my work done, I'm gonna complete the session and my homework.' So I was starting to think, 'Oh, she's helping.'" P7 reported "I don't think this was intended, but sometimes when I felt like I got off track, I would look over at it, and it seemed like it was looking at me being off track, not doing my work, so I was like, okay, I'll get back to doing my work."
最后，学员们表示，学习伴侣让他们有效地管理了学习时间。Blossom 提醒他们合理利用时间，这使他们减少了不必要的休息时间。P1 说："我开始注意到我在想，'我知道在[Blossom]的时间里我能完成很多工作。所以我会想，'好吧，酷。我会坐下来，完成我的工作，完成我的课程和作业。'所以我开始想，'哦，她在帮忙'"。P7 报告说："我不认为这是有意为之，但有时我觉得自己偏离了轨道，我就会看一下它，它似乎在看我偏离了轨道，没有做作业，所以我就想，好吧，我会回去做作业的。

4.2.2 General Feedback. The most common positive feedback from participants related to Blossom's soft, animal-like appearance. The most common criticisms relate to Blossom's loud noises and jerky motions. 2 outlines the major themes from their responses.
4.2.2 一般反馈。参与者最常见的积极反馈与 "花朵 "柔软、动物般的外观有关。最常见的批评意见则与 "花儿 "的大嗓门和生硬的动作有关。2 概述了参与者反馈的主要主题。

Regarding positive feedback that did not directly relate to Blossom's performance during schoolwork sessions, participants frequently cited Blossom's cute, friendly, pet-like appearance, soft, crocheted cover, and small size and things they liked about the robot. P1 stated, "I actually really enjoyed using it. I think it was just cute to have. It was like another presence." P4 stated, "I guess from a
关于与 Blossom 在学校作业中的表现没有直接关系的积极反馈，参与者经常提到 Blossom 可爱、友好、像宠物一样的外观，柔软的钩针编织的外罩，以及小巧的体积，这些都是他们喜欢机器人的地方。P1 说："我真的很喜欢使用它。我觉得它很可爱。它就像另一种存在。P4 说："我想从

Table 2: Positive and negative feedback of the SAR study companion
表 2：叙利亚研究同伴的积极和消极反馈

Table 3: Suggestions and Ideas for the SAR Study Companion
表 3：对《特别研究报告》研究伴侣的建议和想法

visual standpoint, it looks kind of like a pet, so I think that is a pretty good design choice. I think the ears are a good touch."
从视觉角度看，它有点像宠物，所以我认为这是一个很好的设计选择。我觉得耳朵也是一个很好的点缀"。

The most common criticism of the SAR study companion related to Blossom's movements. Participants frequently reported disliking the jerky nature of Blossom's movements, which they found distracting. They also found the noise produced by the robot's servo motors distracting. Participants disliked that the system, including the robot, Raspberry Pi, and touch screen, was too large for some desks and too bulky to be relocated and utilized in other spaces. For example, participants said "the robot is easy to use except that it makes this motor noise when it makes the movements. I found that a bit distracting, but you get used to it if you use it for longer." (P2) and "But yeah, he also takes up a lot of space, so I do have a fairly small desk... All those cables and stuff, it'd be nice to move them around the desk if I wanted to" (P12).
对 SAR 研究同伴最常见的批评与 Blossom 的动作有关。参与者经常表示不喜欢 Blossom 的生涩动作，认为这会分散他们的注意力。他们还发现机器人伺服电机产生的噪音也会分散注意力。与会者不喜欢包括机器人、Raspberry Pi 和触摸屏在内的系统对于某些办公桌来说太大，而且过于笨重，无法搬到其他地方使用。例如，参与者说："机器人很容易使用，但它在做动作时会发出马达声。我觉得这有点让人分心，不过用久了也就习惯了"（P2）和 "不过，是的，我觉得这有点让人分心，不过用久了也就习惯了"。(P2）和 "不过，他也占用了很多空间，所以我的桌子确实比较小......"。所有这些电缆和其他东西，如果我想在桌子上移动的话，那就太好了"（P12）。

4.2.3 Suggested Improvements and Ideas. Participants gave a wide variety of suggestions for system improvements and new functionalities. Table 3 outlines the major themes.
4.2.3 改进建议和想法。与会者就系统改进和新功能提出了各种各样的建议。表 3 列出了主要的主题。

The most common suggestion related to enabling the robot to sense the user's emotional state or periods of distraction and respond. Participants suggested that Blossom could detect when they were distracted and use some cue to recapture their attention. P10 suggested "Maybe it could also incorporate a camera. So the other day I saw that you can capture emotions. Maybe for different emotions, you can give specific voice notes to the person. So like if they are feeling maybe distracted, the robot could say something that could motivate them or something like that."
最常见的建议是让机器人能够感知用户的情绪状态或分心时间，并做出回应。与会者建议，"花朵 "可以检测到用户分心的时间，并使用一些提示来重新吸引他们的注意力。P10 建议说："也许它还可以安装摄像头。前几天我看到可以捕捉情绪。也许针对不同的情绪，你可以给对方特定的语音提示。比如，如果他们感到分心，机器人可以说一些激励他们的话或类似的话。

Conversely, some participants thought that intense monitoring would be unnecessary or unwelcome. P12 stated "When you study with a pet, you know, they can't talk, they're just there and Blossom's just there and I feel like for me that was enough."
相反，一些学员认为，高强度的监控是不必要的或不受欢迎的。P12 说："当你和宠物一起学习时，你知道，它们不能说话，它们就在那里，Blossom 就在那里，我觉得对我来说这就足够了。

When asked to ideate new functions they would find helpful in an in-dorm robot, participants gave many suggestions to improve Blossom and its corresponding system. They suggested adding digital assistant-like features to Blossom, such as reminders, calendaring, assignment tracking, and general AI assistance. P4 stated, "I don't want to say like an AI, like where you ask a question and [the system] answers it, but something similar to that, kind of like Siri or Alexa where you can ask your questions, or maybe do some note-taking things, or even simple things that stop you from looking at your phone, like to check the time or check the weather." P9 suggested interactions to help users consolidate and organize their thoughts as a potential function: "I could just like give it a bunch of ideas and be like, Okay, here's my thought process. Can you just help me out in that way? So, I guess kind of like when you talk to a TA, but like on demand would be really cool."
当被要求提出他们认为对宿舍机器人有帮助的新功能时，与会者提出了许多改进 Blossom 及其相应系统的建议。他们建议在 Blossom 中添加类似数字助手的功能，如提醒、日历、任务跟踪和一般的人工智能协助。P4说："我不想说像人工智能那样，比如你问一个问题，[系统]就会回答，而是类似的东西，有点像Siri或Alexa，在那里你可以问你的问题，或者做一些记笔记的事情，甚至是一些简单的事情，让你不再看手机，比如查看时间或查看天气"。P9 建议将帮助用户整合和整理思绪的交互作为一项潜在功能："我可以给它一堆想法，然后说，好吧，这是我的思考过程。你能以这种方式帮助我吗？所以，我想就像你和助教交谈时那样，但如果能按需提供，那就真的很酷了。"

Regarding the design and appearance of the robot, participants suggested that the touch screen be replaced with a mobile app or physical buttons. P1 suggested "I was more the little screen that I was trying to move. I think if it would be possible, maybe do the interface on a mobile app if that's somehow still connected to her, so it's Blossom herself and not the [touch screen] as well." Others wanted Blossom to be made more portable so that it could be used in other rooms or common spaces. P12 suggested "I think I would really like to just have him as, maybe like a pet; like have him on my desk and then move him over to the kitchen when I'm cleaning, just to be there."
关于机器人的设计和外观，与会者建议用手机应用程序或实体按钮取代触摸屏。P1 建议说："我更喜欢我试图移动的小屏幕。我想，如果可以的话，也许可以在手机应用程序上做一个界面，如果这个应用程序还能以某种方式与她连接的话，那么它就是布卢姆本人，而不是[触摸屏]了"。其他人则希望 Blossom 能够更加便携，以便在其他房间或公共场所使用。P12 建议说："我想我真的很想把它当作宠物来养，比如把它放在我的桌子上，然后在我打扫卫生的时候把它移到厨房，就在那里。

4.2.4 Influence of Recording and Study Procedure. To isolate the effect of the robot from the effect of being recorded, we asked participants if their behavior was impacted by the presence of the webcam and the knowledge that they were being recorded. Three participants ( ) disclosed that Blossom's video and audio recording capabilities made them more aware of their actions and more attentive toward work (P3, P4, P5). Eight participants (73%) reported behaving differently due to being recorded: participants did not use their phones during active sessions ( ), felt added pressure to focus ( ), and suppressed their normal behaviors (such as talking aloud) because they felt selfconscious in front of the camera (P3, P4). However, out of the 10 participants who chose to complete sessions with the robot in condition C , only two opted to close the camera cover during these sessions.
4.2.4 记录和研究程序的影响。为了将机器人的影响与被录制的影响区分开来，我们询问参与者他们的行为是否受到网络摄像头的影响，以及他们是否知道自己被录制了。三位参与者（ ）表示，Blossom 的视频和音频录制功能让他们更清楚自己的行为，也更专注于工作（P3、P4、P5）。八名参与者（73%）表示，由于被录音，他们的行为发生了变化：参与者在活动过程中没有使用手机（ ），感到了更大的压力，必须集中精力（ ），并抑制了自己的正常行为（如大声说话），因为他们觉得在镜头前会有自我意识（P3，P4）。然而，在条件 C 中选择与机器人一起完成游戏的 10 名参与者中，只有两人在游戏过程中选择关闭摄像头盖板。

This work proposed a minimally interactive initial design of an indorm SAR study companion to support college students with ADHD in performing schoolwork tasks. The robot's perceived usability was evaluated through an in-dorm user study spanning multiple weeks with participants sampled from the intended user population, college students with ADHD. Relative to each research question, we found that: 1) participants demonstrated that the system was usable, even with minimal functionality; 2) participants found that
这项研究提出了一个室内 SAR 学习伴侣的最小交互式初步设计，以帮助患有多动症的大学生完成学校作业任务。通过对宿舍用户进行为期多周的研究，对机器人的可用性进行了评估，研究对象是从目标用户群--患有多动症的大学生中抽取的。针对每个研究问题，我们发现1) 参与者认为该系统即使只有最低限度的功能，也是可用的；2) 参与者认为该系统即使只有最低限度的功能，也是可用的。
the robot enacted a sense of companionship and accountability, but found the noise of the robot's motors and jerky movements distracting; and 3) participants gave a variety of suggestions to extend the functionality of the study companion robot. Next, we discuss the results relative to each of the research questions.
3) 参与者对扩展研究陪伴机器人的功能提出了各种建议。接下来，我们将讨论与每个研究问题相关的结果。

The average SUS score of 83.8 and the continued use of the robot when no longer required by all but one of the study participants indicate that our sample of college students with symptoms of ADHD found Blossom useful as a study companion, even with minimal, non-interactive behavior. The single participant who did not start a schoolwork session with Blossom under condition C reported in their post-study survey that they traveled during that week of the study. For this reason, we recommend that researchers hoping to employ a similar structured-to-unstructured study design take care to directly confirm with participants that they will be living in their dorm for the entire duration of the study, as participants may interpret the lack of requirements in the final condition as permission to travel or take other actions that otherwise prevent system use. From the distribution of session start times in Figure 4, we can see that participants rarely started sessions during typical workday hours and instead chose to initiate sessions in the evenings, as late as 1-3 am. One potential interpretation is that Blossom can fill a desire for companionship at times when another person may not be available.
平均 SUS 得分为 83.8 分，而且除一名参与者外，其他所有参与者在不再需要时都会继续使用机器人，这表明我们的样本中有多动症症状的大学生认为 Blossom 作为学习伙伴非常有用，即使它的互动行为极少。在条件 C 下，没有与 Blossom 一起开始学习的唯一一名参与者在他们的研究后调查中表示，他们在研究的那一周里去旅行了。因此，我们建议希望采用类似 "结构化-非结构化 "研究设计的研究人员注意直接向参与者确认他们将在整个研究期间住在宿舍里，因为参与者可能会将最终条件中的无要求理解为允许他们旅行或采取其他行动，从而妨碍系统的使用。从图 4 中的会话开始时间分布可以看出，参与者很少在典型的工作日时间开始会话，而是选择在晚上，甚至凌晨 1-3 点开始会话。一种可能的解释是，Blossom 可以满足人们在他人不在身边时对陪伴的渴望。

When commenting on interactions with the SAR study companion, participants reported that they enjoyed Blossom's subtle movements, yet the jerky nature of the robot's movements and the loud noises that accompanied them were the most common complaints about the interaction. Because participants reported liking the movements, seemingly despite these negative qualities, it is likely that Blossom's idle motions were not inherently distracting to users with ADHD, but that implementing those motions with noisy actuators at high speed or too frequently could be more distracting than helpful. Even with the very loud motors in the present system, participants reported "getting used to" the noises quickly, and, in some cases, the added noise fortified them to stay focused on their work.
在评论与 SAR 研究同伴的互动时，参与者表示他们喜欢 Blossom 的细微动作，但机器人动作的生涩性和伴随的巨大噪音是对互动最常见的抱怨。尽管存在这些负面因素，但参与者还是对机器人的动作表示喜欢，这说明 Blossom 的空闲动作本身并不会分散多动症用户的注意力，但如果使用噪音较大的执行器高速或频繁地执行这些动作，则可能会分散注意力，而非起到帮助作用。即使本系统中的电机声音非常大，参与者也表示很快就 "习惯 "了这些噪音，在某些情况下，额外的噪音还能让他们更加专注于工作。

Based on their feedback, participants appreciated Blossom's zoomorphic appearance, soft exterior, and small size. They generated a wide array of ideas for useful features that could be added to the robot that spanned from those highly related to study companions (e.g., reminders, assignment tracking) to those beyond the problem space of schoolwork (e.g., functionality as a cooking assistant (P6) or coffee-maker (P5). The repeated themes of removing the bulky touch screen and wanting the robot to be more portable inform our immediate next steps in the system design process. We will explore replacing the Raspberry Pi-based UI with a mobile phone app or incorporating a built-in battery pack as a power source. Given the widespread positive feedback regarding Blossom's appearance, we plan to continue to use Blossom in future iterations of the study companion design with the modifications described above.
根据他们的反馈，参与者对 Blossom 的变形外观、柔软的外表和小巧的体型表示赞赏。他们就机器人可添加的实用功能提出了各种各样的想法，其中既有与学习伙伴高度相关的功能（如提醒、作业跟踪），也有超出学校作业问题范围的功能（如作为烹饪助手（P6）或咖啡壶（P5）的功能）。去掉笨重的触摸屏和希望机器人更加便携这些反复出现的主题，为我们下一步的系统设计过程提供了参考。我们将探索用手机应用程序取代基于 Raspberry Pi 的用户界面，或将内置电池组作为电源。鉴于 Blossom 的外观获得了广泛的积极反馈，我们计划在经过上述修改后，继续将 Blossom 用于未来的研究配套设计迭代中。

In all conditions, factors beyond compensation may have motivated them to use the robot when they otherwise would not have done so. Participants may have continued completing daily sessions if they did not see the email reminder to begin condition C or if they intended to "make up" missed requirements for condition B in the previous week. Including a session counter and information about the study condition on the system's UI display would reduce some of the confounds about participant motivation. Experimenter demand effects [20,37] may have also played a role in the observed results; participants may have used the robot with the belief that the researchers wanted or expected them to do so. One way to limit this effect in future studies is to refrain from video- and audio-recording the sessions to minimize participants' beliefs that the researchers will know when they have not used the system.
在所有条件中，都可能有补偿以外的因素促使他们使用机器人，否则他们就不会这样做。如果参与者没有看到开始条件 C 的电子邮件提醒，或者他们打算 "弥补 "上一周错过的条件 B 的要求，那么他们可能会继续完成每天的课程。在系统的用户界面显示中加入会话计数器和研究条件信息，可以减少一些与参与者动机有关的干扰因素。实验者的需求效应[20,37]也可能对观察到的结果产生影响；参与者可能认为研究人员希望或期望他们这样做，从而使用了机器人。在未来的研究中，限制这种效应的方法之一是避免对研究过程进行录像和录音，以尽量减少参与者认为研究人员会知道他们没有使用系统的想法。

Although our study provided preliminary insights about the viability of study companion robots, future work should investigate how the robot's behavior affects the study session interaction. Future research should explore a condition in which the robot is present but does not display any movement to determine whether the voluntary use observed in our study can be attributed to the robot's idle motions. Finally, future work will include the analysis of the study session recordings. In line with participants' suggestions to equip the robot with user monitoring capabilities, we intend to analyze visual and audio features to understand how participants' behavior differed between conditions and how multi-modal data can be used to predict the user's state during a study session.
虽然我们的研究为学习陪伴机器人的可行性提供了初步见解，但未来的工作应该研究机器人的行为如何影响学习过程中的互动。未来的研究应探讨机器人在场但不显示任何动作的情况，以确定在我们的研究中观察到的自愿使用是否可归因于机器人的空闲动作。最后，未来的工作将包括分析研究会话录音。根据参与者提出的为机器人配备用户监控功能的建议，我们打算分析视觉和音频特征，以了解参与者在不同条件下的行为有何不同，以及如何利用多模态数据来预测用户在学习过程中的状态。

This work contributes the participatory design and evaluation of a socially assistive robot study companion for college students with ADHD. Our findings show that: RQ1) A sample of college students with ADHD found the system useful, even in its initial pilot state, and elected to continue using it when they were no longer required to; RQ2) College students with ADHD have demonstrated use for and interest in in-dorm SAR study companions, although users have differing and unique preferences for the robot's behavior in this role; In addition, users found the loud noises and abrupt movements of the robots' servomotors distracting during a study session, but they were able to acclimate with repeated use; and RQ3) test users proposed many ideas for potential improvements to the robot's design and additional features that they would find useful in an in-dorm robot. These findings suggest that in-dorm robots have potential as long-term assistive devices for college students with ADHD. Furthermore, we demonstrated the feasibility of incorporating a long-term in-dorm deployment into the early stages of a participatory design process for Human-Robot Interactions.
这项工作有助于为患有多动症的大学生设计和评估一款社交辅助机器人学习伴侣。我们的研究结果表明问题 1）有多动症的大学生样本发现该系统非常有用，即使是在最初的试验状态下，当他们不再需要使用该系统时，也会选择继续使用它；问题 2）有多动症的大学生表现出对宿舍内机器人学习伙伴的使用和兴趣，尽管用户对机器人在这一角色中的行为有不同和独特的偏好；此外，用户在学习过程中发现机器人伺服电机发出的巨大噪音和突然的运动会分散他们的注意力，但反复使用后他们就能适应了；以及问题 3）测试用户提出了许多改进机器人设计的潜在想法，以及他们认为对宿舍内机器人有用的其他功能。这些研究结果表明，宿舍机器人有可能成为多动症大学生的长期辅助设备。此外，我们还证明了将长期宿舍内部署纳入人机互动参与式设计流程早期阶段的可行性。

This research was supported by the National Science Foundation Grant NSF IIS-1925083 and the NSF REU program. The authors extend additional thanks to Caroline Kenney and Anna-Maria Valentza.
本研究得到了美国国家科学基金会 NSF IIS-1925083 基金和美国国家科学基金会 REU 项目的支持。作者还要感谢 Caroline Kenney 和 Anna-Maria Valentza。