AI-Based System to Increase Productivity by Detecting Attention Span
Disciplines
Computer and Systems Architecture | Robotics
Abstract (300 words maximum)
Distraction and cognitive fatigue erode productivity in labs, classrooms, and offices, yet most environments provide only static cues (fixed lighting, generic reminders) that fail to adapt to an individual’s moment-to-moment state. This project proposes an AI-driven, multimodal sensing system that estimates attention in real time and delivers gentle, context-aware feedback to help users sustain focus without becoming intrusive. The system combines short-range depth (time-of-flight) vision with a compact thermal module and ambient sensors (light, temperature, acoustics), extracting non-identifying indicators such as blink rate and stability, gaze drift, head pose, and micro-motions, yawning frequency, posture shifts, noise spikes, and comfort deviations. Features aim to be fused on-edge using a lightweight classification model with adaptive modality weighting so that the most reliable signals in each setting (e.g., gaze under good illumination or posture when lighting degrades) drive the estimate. When a rising distraction score persists beyond a brief window, the system triggers tailored nudges: dimming or brightening task lighting toward a comfort target, proposing a micro-break or breath cycle, suppressing noncritical notifications, or prompting small ergonomic adjustments; all actions are logged locally for user review and tuning. Privacy is prioritized through on-device processing, ephemeral buffering, and no facial identification. The evaluation plan uses a within-subject design comparing baseline sessions to assistance-enabled sessions on timed focus tasks, measuring changes in sustained attention proxies (task completion time, error rate), interruption recovery, and self-reported workload and comfort, alongside stability of the model’s estimates across lighting and noise variations. Initial results seek to demonstrate dependable detection of common distraction patterns, timely and acceptable feedback that users do not find disruptive, and measurable improvements in task continuity and perceived effort. Future extensions include few-shot personalization to new users, adaptive schedules that anticipate circadian dips, and opt-in integration with productivity tools to coordinate breaks and deep-work windows.
Use of AI Disclaimer
no
Academic department under which the project should be listed
SPCEET – Robotics and Mechatronics Engineering
Primary Investigator (PI) Name
Ravzan Voicu
AI-Based System to Increase Productivity by Detecting Attention Span
Distraction and cognitive fatigue erode productivity in labs, classrooms, and offices, yet most environments provide only static cues (fixed lighting, generic reminders) that fail to adapt to an individual’s moment-to-moment state. This project proposes an AI-driven, multimodal sensing system that estimates attention in real time and delivers gentle, context-aware feedback to help users sustain focus without becoming intrusive. The system combines short-range depth (time-of-flight) vision with a compact thermal module and ambient sensors (light, temperature, acoustics), extracting non-identifying indicators such as blink rate and stability, gaze drift, head pose, and micro-motions, yawning frequency, posture shifts, noise spikes, and comfort deviations. Features aim to be fused on-edge using a lightweight classification model with adaptive modality weighting so that the most reliable signals in each setting (e.g., gaze under good illumination or posture when lighting degrades) drive the estimate. When a rising distraction score persists beyond a brief window, the system triggers tailored nudges: dimming or brightening task lighting toward a comfort target, proposing a micro-break or breath cycle, suppressing noncritical notifications, or prompting small ergonomic adjustments; all actions are logged locally for user review and tuning. Privacy is prioritized through on-device processing, ephemeral buffering, and no facial identification. The evaluation plan uses a within-subject design comparing baseline sessions to assistance-enabled sessions on timed focus tasks, measuring changes in sustained attention proxies (task completion time, error rate), interruption recovery, and self-reported workload and comfort, alongside stability of the model’s estimates across lighting and noise variations. Initial results seek to demonstrate dependable detection of common distraction patterns, timely and acceptable feedback that users do not find disruptive, and measurable improvements in task continuity and perceived effort. Future extensions include few-shot personalization to new users, adaptive schedules that anticipate circadian dips, and opt-in integration with productivity tools to coordinate breaks and deep-work windows.