[AfA2026-PhysicalAI] - Smart Blind Stick 

Modulino™ Distance

Speaker (generic)

USB Hub, Bus Powered

Arduino App Lab

Webcam, Logitech® HD Pro

Arduino® UNO™ Q 2GB

<html><body>The Smart Blind Stick is an advanced assistive device designed to enhance the independent mobility of individuals with visual impairments. By bridging high-performance computing with real-time hardware control, this system provides "auditory sight" to the user. Using the Arduino UNO Q, the stick identifies people in the environment via a local AI Object Detection model while simultaneously calculating precise distance measurements using a Time-of-Flight sensor. When a person is detected, the device uses a Text-to-Speech (TTS) engine to announce the object's presence and its exact distance in centimeters, providing a real-time verbal map of the surroundings. <h3>Hardware Lineup</h3><h3></h3><figure><img src="https://projects.arduinocontent.cc/7a7237d4-3afa-471c-a2e8-5e7a88cd5834.jpg"/><figcaption></figcaption></figure> Core Controller: Arduino UNO Q featuring a Qualcomm Dragonwing™ QRB2210 MPU and an STM32U585 MCU. Vision Input: USB Camera connected to the MPU for real-time video inference. Distance Sensor: Modulino Distance sensor (Time-of-Flight) connected via the I2C interface for depth sensing. Feedback Output: External speaker system connected to a local TTS server for voice announcements. Connectivity : Integrated Wi-Fi and Bluetooth for communication between the board and the voice synthesis server. <h3>User Interface</h3>This device utilizes a non-visual interface to ensure maximum accessibility and utility for the user: Auditory Feedback : This is the primary interaction method. The system translates technical AI metadata and sensor data into natural language through TTS (Text to Speech). The feedback can be given through a speaker (Bluetooth) or headphones (wired).Spatial Awareness : The system provides specific verbal cues, such as "I see a person at 120 centimeters," allowing the user to judge depth and proximity instantly without manual physical probing. <h3>AI Model</h3>The system employs the YOLOX-Nano object detection model, integrated as a modular software component. Local Edge AI: All AI inference is performed locally on the Debian-based MPU, ensuring user privacy and low-latency response times. Information Prioritization: To ensure the auditory stream remains clear and actionable, the system uses a 0.6 confidence threshold and a 5-second alert cooldown. This prevents the user from being overwhelmed by repetitive or low-confidence notifications, focusing only on significant environmental changes. <h3>Software Architecture</h3>The application logic is distributed across a specialized dual-system setup to ensure reliability: MCU : The STM32 microcontroller runs on Zephyr OS, managing the Modulino Distance sensor and notifying the processor of new measurements every 100ms via the RPC Bridge. MPU : The Qualcomm processor runs Debian Linux and the Python-based application logic. It manages the video stream, executes the object detection model, and coordinates TTS triggers. Communication : The MCU and MPU communicate via Remote Procedure Calls (RPC), allowing the MCU to trigger Python functions as if they were local commands. TTS Infrastructure: A secondary Python Flask server receives text strings over the network and synthesizes them into speech using the pyttsx3 library. <h3>A Quick Look into the Code</h3>In this section, we’ll focus on the flow of detecting an object and triggering a voice alert. The Processor: Detecting a Person On the Processor side (main.py), the logic is event-driven. We use a handler to listen for detections from the camera stream. When a person is recognized with high confidence, the script combines this with the latest distance data and triggers the voice engine: <pre># Function to handle detections and trigger voice </pre><pre>def handle_detections(detections): </pre><pre>	 for key, value in detections.items(): </pre><pre>	 if value.get("confidence") &gt; 0.6: </pre><pre>	 # Construct message with the latest distance from the MCU </pre><pre>	 message = f"I see a {key} at {current_distance} millimeters" </pre><pre>	 speak_text(message) # Send to TTS Server </pre><pre># Connect the callback to the detection stream </pre><pre>detection_stream.on_detect_all(handle_detections) </pre>The MCU: Reading Distance On the Microcontroller side (sketch.ino), the board constantly polls the sensor. It sends the raw distance data to the processor using the Bridge: <pre>void loop() { </pre><pre>	 if (distanceSensor.available()) { </pre><pre>	 int measurement_mm = distanceSensor.get(); </pre><pre>	 </pre><pre>	 // Notify the Linux processor of the new distance </pre><pre>	 Bridge.notify("update_distance", measurement_mm); </pre><pre>	 } </pre><pre>} </pre><h3>Visuals &amp; Media</h3><h3></h3> <figure><iframe src="https://youtu.be/_bByYzJ9ezA"></iframe></figure> <figure><img src="https://projects.arduinocontent.cc/34ca2e69-ce03-4d05-ae91-5223baeeadd9.jpg"/><figcaption></figcaption></figure> <div id="gallery"><img id="gallery" src="https://projects.arduinocontent.cc/3755f9ed-7628-4a33-a9ad-7961eb14e00c.jpg"/><figcaption></figcaption></div> <h3></h3><h3></h3> <h3></h3> </body></html>

[AfA2026-PhysicalAI] - Smart Blind Stick

An AI-powered smart blind stick that uses on-device computer vision and distance sensing to provide real-time spoken navigation assistance for visually impaired users.