Owner-Following Voice-Verified Puppy Robot

ISEF Category: Robotics and Intelligent Machines

Subcategory: Other  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

A robot can hear you, but should it trust you? That question matters when a home robot uses your voice to decide whether to come closer. If it also tracks you with a UWB tag, then small sensing errors can turn into big behavior mistakes. Your project tests how well that combo works in a real, messy home.

What Is It?

This project combines two ideas. The first is speaker verification, which checks whether a voice matches a stored owner voice pattern. The second is UWB tracking, which uses short radio pulses to estimate distance and direction with a tag, like a digital leash that does not need cameras.

Think of the robot like a puppy with two senses. Hearing tells it who is calling. The UWB tag tells it where to go next. Your job is to see how well those senses work together when the room is cluttered, when different people call the robot, and when furniture blocks clean movement.

That mix makes the project strong. You are not just building a cute robot. You are testing how real home conditions affect trust, navigation, and follow accuracy.

Why This Is a Good Topic

This is a strong science fair topic because you can test a clear system with measurable outputs. You can change one factor at a time, like caller identity, room clutter, or tag position, and measure response time, path error, and false acceptance. The project connects to home robotics, assistive tech, and safe human-robot interaction. You can also learn signal processing, basic machine learning, motion planning, and experimental design.

Research Questions

• How does room clutter affect the robot’s path error when it follows the owner’s UWB tag? ?
• What is the effect of different speakers on the voice-verification false accept rate? ?
• Does the robot respond faster to the owner than to non-owners across repeated trials? ?
• To what extent does changing tag placement on the body change following accuracy? ?
• Which home layout produces the largest gap between voice recognition accuracy and real following accuracy? ?
• How does background noise change the robot’s ability to trigger the correct follow behavior? ?

Basic Materials

• Small mobile robot chassis with motor driver and controller board.
• Microphone module with on-device audio input support.
• Low-power speaker-verification model setup on an edge device, such as a Raspberry Pi or similar single-board computer.
• UWB tag and compatible DWM3000 or similar UWB anchor hardware.
• Battery pack and power regulators for the robot and sensing modules.
• Tape measure for ground-truth distance checks.
• Painter’s tape for marking paths and start points.
• Household obstacles such as boxes, chairs, and laundry baskets for clutter tests.
• Smartphone or tablet for recording trial videos.
• Notebook or spreadsheet for logging trial outcomes.

Advanced Materials

• Single-board computer or embedded AI module with enough compute for onboard inference.
• Multiple UWB anchors for improved location estimates.
• Spectrum or audio analysis tools for voice-signal inspection.
• Motion capture or overhead camera setup for ground-truth tracking.
• ROS-compatible robot base for logging navigation data.
• IMU sensor for comparing heading estimates with UWB position estimates.
• Calibrated reference microphone for checking audio quality across trials.
• 3D-printed mount for keeping the microphone and tag in fixed positions.

Software & Tools

• Python: Organizes trial data, runs statistics, and plots accuracy versus clutter level.
• OpenCV: Tracks the robot’s path from video so you can compare planned and actual motion.
• pandas: Cleans tables of trial results and groups them by owner, room, and obstacle level.
• ImageJ: Measures distances and angles from saved frames when you need a quick visual check.
• ROS: Records robot motion, sensor data, and control events in a structured way.

Experiment Steps

1. Define the exact success criteria for a valid follow event, including voice trigger, tag lock, and path completion.
2. Choose one independent variable first, such as caller identity or clutter level, so you can isolate the main failure point.
3. Build a ground-truth method for measuring follow accuracy, such as floor markers, video tracking, or logged position data.
4. Plan control trials that separate voice-verification errors from navigation errors, so you know which subsystem caused each miss.
5. Create a data table that records response time, false triggers, path deviation, and trial conditions in the same format every time.
6. Decide how you will compare owners and non-owners with the same statistical test across all trial sets.

Common Pitfalls

• Training the voice model on too little audio, which makes the robot accept the wrong person during testing.
• Placing the UWB tag differently from trial to trial, which changes signal quality and hides the real effect of clutter.
• Measuring success only by whether the robot started moving, which misses path error and owner-follow mismatch.
• Testing in rooms with changing background noise, which makes audio-trigger results hard to compare across sessions.
• Using too few trials per condition, which makes random navigation wiggles look like a real effect.

What Makes This Competitive

A stronger version of this project will separate sensing from behavior. You can compare voice-only, UWB-only, and combined control, then measure where each one fails. You can also test several owners, several room layouts, and a harder metric than simple success rate, such as path deviation or time-to-reacquire after signal loss. That turns the project from a demo into a careful study of human-robot trust and reliability.

Project Variations

• Test how well the same robot follows different owners in homes with open floors versus heavy furniture.
• Swap the voice model for a simpler keyword detector and compare owner-specific trigger accuracy against speaker verification.
• Keep the voice system fixed, then compare UWB tracking performance when the tag sits in a pocket, on a wrist, or on a collar.

Learn More

• NIH PubMed: Search review articles on speaker verification, UWB localization, and human-robot interaction to find background and methods.
• NASA Tech Reports Server: Search for papers on indoor navigation, sensor fusion, and robot localization methods that use radio ranging.
• MIT OpenCourseWare: Look for free robotics, signal processing, and machine learning course materials to build your technical background.
• IEEE Xplore abstracts: Search for recent papers on ECAPA-TDNN, UWB positioning, and household robot following, then read the abstracts and figures where full text is not available.
• ROS Wiki: Read the free documentation for robot navigation, sensor integration, and logging if you plan to use ROS.