Cognitive Load Notification Robot

ISEF Category: Robotics and Intelligent Machines

Subcategory: Cognitive Systems  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

Your brain gives off tiny clues when you are overloaded. You blink differently, move your head less, and type in a new rhythm. A desk robot can read those clues and decide when to stay quiet. That turns a noisy notification problem into a smart timing problem.

What Is It?

This project asks whether a robot can tell when you are mentally busy and hold notifications until you can handle them better. The robot watches signals from a webcam and keyboard, then makes a decision about whether to interrupt you. In plain language, it is trying to learn when your brain is "full."

Think of it like a careful friend who waits for a pause before asking a question. Blink rate, small head movements, and typing cadence can act like traffic signals for mental effort. When those signals suggest high load, the system can delay a pop-up or message. When the signals suggest a lighter moment, it can allow the interruption. Your job is to test whether that timing helps people finish tasks faster or with fewer mistakes.

Why This Is a Good Topic

This is a strong science fair topic because you can test a clear claim, adaptive interruptions can improve performance compared with random or always-on notifications. You also get a real-world link to study habits, screen distraction, and human-computer interaction. A student can learn data collection, model evaluation, and experiment design without needing to invent new hardware from scratch.

Research Questions

• How does adaptive notification timing based on blink rate affect task completion time?
• What is the effect of using head micro-movements as a load signal on interruption accuracy?
• Does typing cadence predict when a user should receive a notification better than random timing?
• To what extent does a contextual-bandit policy reduce missed task steps compared with an always-on baseline?
• Which combination of webcam and typing features best separates high-load and low-load moments?
• How does notification gating affect task errors during sustained attention tasks?
• What is the effect of different task types on the reliability of cognitive load detection?

Basic Materials

• Laptop or desktop computer with webcam and keyboard input.
• External mouse or trackpad.
• Screen-recording software.
• Basic task software for reading, typing, or puzzle tasks.
• Consent form for any human participants.
• Timer or stopwatch app.
• Spreadsheet software for logging results.

Advanced Materials

• High-resolution webcam with stable mounting.
• Optional eye-tracking or face-landmark sensor setup.
• Microphone for detecting notification response timing.
• Python development environment.
• Machine learning libraries for feature extraction and policy testing.
• Video annotation software.
• Statistical analysis software.

Software & Tools

• Python: Processes webcam and typing data, then runs feature extraction and model testing.
• OpenCV: Detects facial landmarks, blink events, and head motion from video frames.
• ImageJ: Helps inspect video frames and measure movement or signal quality by hand.
• R: Runs statistical tests and compares notification policies across task conditions.
• JASP: Lets you test group differences and build clean graphs without paid software.

Experiment Steps

1. Define the user task you will protect from interruption, and decide what counts as success.
2. Choose the signals you will measure first, then set a clear rule for how they represent cognitive load.
3. Build a baseline that sends notifications without adaptation, so you have a fair comparison.
4. Design a simple decision policy, then decide how it will switch between interrupt and wait states.
5. Plan controls that separate notification timing effects from task difficulty and user differences.
6. Map out how you will score speed, errors, and response timing, then choose the statistics that match your design.

Common Pitfalls

• Training on too few people, which makes the load detector memorize one typing style instead of learning a general pattern.
• Using a webcam angle that hides the face, which makes blink rate and head motion measurements noisy.
• Mixing task difficulty with notification timing, which makes it hard to tell whether the robot or the task caused the result.
• Calling every pause a low-load moment, which can confuse focused thinking with distraction.
• Comparing policy performance only by speed, which misses accuracy loss and weakens the conclusion.

What Makes This Competitive

A stronger project would test the system on more than one task type and report both speed and accuracy, not just one outcome. You would also want a clear baseline, a real decision policy, and evidence that the model works on new users, not only the people in your training set. Better still, compare simple rules against the contextual-bandit approach to prove the adaptive method adds value. Clean statistics and careful controls matter as much as the code.

Project Variations

• Test whether the same cognitive load model works during reading, typing, and logic puzzles instead of only one task.
• Replace the contextual-bandit policy with a simple rule-based notification system and compare the results.
• Use only typing cadence, only blink rate, or only head motion to see which signal carries the most predictive power.

Learn More

• MIT OpenCourseWare, Human-Computer Interaction materials: Search MIT OpenCourseWare for courses on HCI, attention, and interactive system design.
• NIH PubMed: Search review articles on cognitive load, interruption management, and human factors in digital work.
• NASA Human Factors: Find articles and reports on attention, workload, and operator performance in complex systems.
• OpenCV documentation: Read the free docs for face detection, landmark tracking, and video analysis.
• JASP: Download the free statistics app to compare notification conditions and summarize participant results.