Hapster: Apple Watch Haptic Feedback
- Hapster is a prototype student response system that uses Apple Watch haptics to provide live, aggregated, anonymous feedback in physical classrooms.
- It integrates a student web interface, an instructor watchOS app, and a researcher dashboard, employing REST API, Firebase Firestore, and aggregation rules to streamline real-time communication.
- The evaluation reveals enhanced instructor awareness and student engagement, while highlighting challenges like haptic differentiation and managing reaction spamming.
Searching arXiv for the cited Hapster paper and related entries. arXiv search query: Hapster Apple Watch haptics classroom student feedback Hapster is a prototype student response system (SRS) for in-person classrooms that uses an Apple Watch to deliver live, aggregated student feedback to instructors through both visual and vibrotactile notifications. It was designed to address a limitation of existing classroom feedback tools such as Slido or Mentimeter: these systems typically require instructors to look at a screen to notice student input, which is inconvenient or impossible when the instructor is writing on a board, moving around, or otherwise not facing the display. Hapster therefore explores whether haptics can serve as a practical live feedback mechanism in a physical classroom while preserving low-friction, anonymous student participation and multimodal instructor awareness (Golev et al., 8 Jul 2025).
1. Design objective and conceptual scope
Hapster was built to let students send simple, anonymous reactions during lecture and let the instructor receive them live in a way that is low-friction for students, anonymous to reduce social pressure, aggregated to reduce noise, and multimodal for instructors, combining a watch visual log with haptic alerts. The motivating premise is that immediate feedback can improve learning and anonymity increases student willingness to participate, but visual-only SRSs still force instructors to split attention. The system accordingly treats touch, rather than a conventional display, as the primary notification channel.
The platform is explicitly oriented toward the physical classroom. Its core use case is lecture delivery under conditions in which instructors may be writing on a blackboard, annotating projected material, or moving through the room. In that setting, Hapster aims to make student reactions immediately perceptible through touch, so instructors can notice them without diverting their eyes. The paper frames this as a response to the absence of prior classroom SRSs that had explored haptics as the main notification channel (Golev et al., 8 Jul 2025).
A plausible implication is that Hapster is not merely a variant of existing polling software, but a reformulation of the instructor-notification problem. The emphasis shifts from collecting responses to making those responses perceptible under attentional constraints characteristic of live teaching.
2. System architecture and interaction flow
Hapster has three main front ends plus a backend. The instructor-facing watchOS app runs on an Apple Watch, lets the instructor create and end a session, displays incoming student reactions as a live log, and triggers a unique haptic sequence for each reaction type. The student-facing web app runs in any browser-compatible device; students enter a session ID, choose one of three reaction buttons, and send the submission anonymously to the instructor’s watch. A researcher-facing dashboard is used during studies for live monitoring and shows live reaction statistics and timelines. The backend is a REST API written in FastAPI; it uses Firebase Firestore to store session data at session termination, the web components are built with React and deployed on Render, and the watch app is built with SwiftUI and distributed via TestFlight (Golev et al., 8 Jul 2025).
The instructor workflow is minimal. The instructor opens the watch app and taps Begin Session; the app generates a random session ID; the instructor shares the ID with the class; as students react, the watch shows them in a visual log with an emoji and label and also plays a haptic sequence for each reaction; the instructor can tap End Session when done. The student workflow is similarly compressed: students open the web app, enter the session ID, and are shown three reaction buttons. When a button is tapped, the student sees a short confirmation animation with the corresponding emoji, the input is transmitted to the instructor’s watch, and a 20-second cooldown disables that reaction button to discourage spamming.
This architecture makes the system simultaneously a classroom tool and a research instrument. The researcher-facing dashboard, session-level storage, and live monitoring functionality indicate that the prototype was designed not only for deployment but also for controlled observation of reaction streams, aggregation behavior, and instructor response.
3. Feedback vocabulary, haptic encoding, and aggregation
The supported feedback vocabulary consists of three student reactions selected from an instructor survey using a modified Borda count from a pre-survey of 26 instructors teaching large lectures. The top three were Confused / frustrated with 37 points, Hand-raising for questions/comments with 29 points, and Confident in understanding with 21 points. These three reactions were identified as the most desired across instructors (Golev et al., 8 Jul 2025).
| Reaction | Meaning | Haptic mapping |
|---|---|---|
| Hand-Raise | The student has a question/comment | WKHapticType.notification, described as feeling like a heartbeat |
| Confused | The content is confusing or frustrating | WKHapticType.failure, described as a strong vibration |
| Confident | Understanding / readiness to move on | WKHapticType.directionUp, described as two short ticks |
A central design choice is that each reaction maps to a specific Apple Watch haptic type, and each watchOS haptic pattern is played four times in quick succession to create a recognizable sequence. The rationale was to make the more important or intrusive reactions, especially Confused and Hand-Raise, use stronger, more attention-grabbing haptics, while Confident was mapped to a lighter pattern. Apple Watch was chosen because it is widely adopted, supports vibrotactile notifications, and its predefined haptic library was sufficient for exploratory use. The paper notes that watchOS does not allow customizable vibration frequency, but this was not considered a major issue because people are not highly sensitive to frequency; the implementation instead relies on differences in amplitude, duration, and perceived feel of the patterns.
Hapster does not treat every student input as a separate notification when they occur close together. Repeated reactions of the same type within a 10-second window are aggregated: they are grouped together, the watch displays the reaction with an aggregated count, and the corresponding haptic sequence plays only once. The researcher-facing dashboard likewise aggregates data as a reverse-chronological reaction log, a line graph showing reactions over the last 10 minutes, a bar chart of cumulative reaction counts, and a pie chart of counts per anonymous user. This aggregation rule is procedural rather than formalized by an explicit equation, but it is central to the system’s strategy for reducing notification noise during active lecture periods.
4. Evaluation protocol and study structure
The evaluation was conducted in a multi-stage field study at a U.S. university. A pre-survey of 26 instructors ranked 12 candidate student reactions, from which the top three were selected using the modified Borda count. A pilot study in a computer science lecture with 60 students found no technical issues, but the instructor reported that the haptics should be more intrusive and longer to notice while lecturing; this led the team to repeat the haptic patterns more times in the sequence. The main in-class study involved 6 instructors, 2 female and 4 male, and 155 students total across 6 lectures. Classes lasted 50 or 80 minutes and included blackboard-based instruction, projected iPad writing, projected slides, different disciplines, and different enrollment sizes. The six classes included C1 with 43 students present, C2 with 200 enrolled and 42 active participants observed in study context, C3 with 60 students present, C4 with 90 students present, C5 with 50 students present, and C6 with 70 students present (Golev et al., 8 Jul 2025).
Before each lecture, instructors received the Apple Watch, were trained on the three haptic sequences, heard each sequence played twice, were asked to describe them, and then completed a recognition test of 9 sequences total, corresponding to 3 sequences multiplied by 3 repetitions in random order. After each session, instructors were interviewed, students completed a survey, and both groups completed NASA-TLX workload questionnaires. The student survey covered lecture experience, comparison to other SRSs, and perceived impact on instructor behavior. Of 72 student survey responses collected, 7 were excluded for non-consent and 3 were excluded for referencing an untested class, leaving 62 valid responses. Student participation was optional and had no grade penalty; instructors did not know which students used the system; researchers observed in person but did not record video or audio.
The study design therefore combines deployment realism with structured evaluation. It captures classroom heterogeneity, instructor training effects, post-session reflection, and workload assessment, while preserving the anonymity conditions central to the system’s conceptual motivation.
5. Empirical results
The system was generally seen as effective at delivering live feedback. Students reported that Hapster let them communicate more than they otherwise would, and the paper concludes that the system was effective for enabling anonymous student communication. Student survey responses indicated that Hapster let them convey more information than without the system with , made them more willing to interact in the moment with , and left them comfortable using it with . Students also agreed that the system was visually appealing with , while they were more neutral about whether it made them feel more engaged with (Golev et al., 8 Jul 2025).
Instructors reported that Hapster increased their awareness of student reactions, made communication feel more bi-directional, and helped them adjust pacing live. Five of six instructors said the feedback helped them slow down when confused reactions appeared and speed up when confident reactions accumulated. One instructor noted that seeing confidence reactions was reassuring, and another said it was especially useful for first-time teaching of a course, where timely feedback can help improve the curriculum.
Haptic perception results were mixed but informative. Five of six instructors memorized and recognized the three sequences. All instructors except one identified Hand-Raise with 100% accuracy in the post-training test. Confident was generally easiest to distinguish, whereas Confused was often mistaken for Hand-Raise or Confident. The normalized confusion matrix showed that Hand-Raise was best recognized, Confident was moderately recognized, and Confused was the hardest. A key issue was that Hand-Raise and Confused felt too similar to several instructors. One instructor, despite training, could not distinguish any of the three haptics during lecture, found the vibrations distracting, and said they could not sense the difference; that instructor did not engage with the watch during class, and students consequently judged the system’s effectiveness as inconclusive. This finding places clear weight on haptic sensitivity and on the instructor’s willingness to attend to the watch while lecturing.
Workload results reinforce the same pattern. Students found the system lightweight, reporting mental demand of and physical demand of . Instructors also rated physical demand low at , but reported higher mental demand at . This suggests that the cost of the system is borne less in physical interaction than in the cognitive effort of lecturing while simultaneously interpreting haptic cues.
6. Tradeoffs, limitations, and prospective directions
A major theme is the tension between flexibility and abuse potential. In some sessions, a small number of students spammed reactions; in one class of 200, one student produced 30% of all reactions, and some students kept submitting reactions every 20 seconds even with cooldowns. This created situations in which instructors received repeated Confused or Hand-Raise signals even when no one actually asked a question. The paper therefore presents stronger moderation as an unresolved systems problem rather than a peripheral usability issue (Golev et al., 8 Jul 2025).
Possible mitigation strategies mentioned include limiting the number of reactions per unique user per session, warning and then banning users who exceed a threshold such as 4 reactions per 5 minutes, allowing instructors to choose their own reaction set and haptic mappings, and varying haptic intensity based on the aggregated count. These proposals indicate that the aggregation mechanism alone is insufficient as an anti-abuse control.
The paper also explicitly identifies accessibility as requiring more exploration. It notes that haptics may work differently depending on sensory sensitivity, that success varies by individual, and that additional modalities may be needed for users with different physical, mental, or sensory capabilities. It suggests integrating other wearables or interaction modes to improve accessibility and also proposes that a physical clicker-type device without a visual interface might be easier for students than using a personal electronic device and could reduce distractions.
The stated limitations are a small number of instructor participants, one instructor unable to distinguish haptic sequences, sequences that may not be distinct enough, student spamming despite cooldown, restriction to three reaction types, and dependence on Apple Watch’s limited haptic palette. The paper further notes that a more programmable, multi-actuator wrist device could enable better control over activation patterns, frequency, amplitude, rhythm, waveform, and duration. Future research directions correspond directly to these constraints: more distinct haptic design, exploration of multi-actuator devices, customizable reactions, intensity proportional to aggregation, better anti-abuse mechanisms, alternative student interfaces, and broader multimodal accessibility research.
Taken together, these results position Hapster as a multimodal classroom SRS whose principal contribution lies in testing whether live haptic feedback can reduce the attentional burden of visual-only classroom response systems. The empirical record presented is positive but conditional: haptics can function as an effective live feedback mechanism for instructors in the physical classroom, but success depends on sequence design, instructor perceptual fit, and robust moderation.
