Do Not Disturb Robotics Arm Project

My feelings to this project

This project means everything to me. I built it solo during a four-day hackathon, fueled by passion and just six hours of sleep. I was so exhausted afterward that I bombed my finals in statistics and signal processing. You could say I traded my GPA for this project, but it was worth it.

That was the moment I knew I had to be in robotics—it just clicked. It was the perfect fusion of computer vision, embedded systems, control theory, and PCB design, all working together to bring a complex robotic arm to life. I still vividly remember the late nights: writing error handling / testing scripts, recording and adding safety angular constraints,troubleshooting communication failures on the shared serial port, and fixing the burned out motors from overcurrent.

The more I thought about it, the more I understood. This world may not be a world of science; it might be a complex world that are engineered with 1000 more preprocessing steps.

Introduction

The “Do Not Disturb” project is an integrated robotics system designed to intelligently prevent interruptions based on a user’s mental state. The project was conceived from the personal experience of being disrupted while in a state of deep focus, or “flow.” By combining Brain-Computer Interface (BCI) technology with computer vision and a robotic arm, the system serves as an automated guardian that only activates when the user is genuinely concentrating, solving the nuanced problem of distinguishing between welcome and unwelcome interruptions.

Why This Project and What It Does

The primary motivation behind this project was to create a system that could understand the user’s level of focus and act accordingly. The creator identified that not all interruptions are undesirable; sometimes, a distraction is welcome when one is not deeply engaged in a task. The challenge was to build a system that could differentiate between a focused state and an unfocused one.

The system operates through the collaboration of two main components:

1. Brainwave Analysis System: A user wears a BCI headset that captures EEG (electroencephalogram) data. This raw data is first processed by the manufacturer’s API and then fed into a custom deep learning network. This network, trained on public data, performs a two-category classification to determine if the user is currently “focusing” or “not focusing.”
2. Robotic Guardian System: The classification result is sent over the cloud to a Raspberry Pi that controls a robotic arm. If the system detects the user is in a state of focus, it activates the arm. Using an onboard camera, the arm employs a visual tracking algorithm to detect and follow the face of any person who approaches, acting as a clear, non-verbal “Do Not Disturb” signal.

Technical Details

The project’s architecture integrates complex software and hardware components, primarily developed in Python for its rapid prototyping capabilities and SDK availability.

• Computer Vision: The visual tracking system uses a two-stage approach for robust performance. It first uses Haar Cascades for general human face detection. Then, a deep learning model is used for more precise identification and tracking. To prevent erratic arm movements, a “lockdown period” was implemented; if the target face is lost, the system waits a few seconds and predicts the face’s next likely position before re-engaging, ensuring smoother motion.
• Robotic Arm Control: The robotic arm has **2 degrees of freedom ** and is controlled using a PID (Proportional-Integral-Derivative) algorithm. This required the meticulous tuning of six distinct PID parameters (three for each axis) to accurately translate the 2D coordinates from the camera feed into physical movement. This process was a significant challenge, involving extensive debugging of signs and values to correct initial bugs that caused the arm to lock up or behave erratically.
• System Integration and Challenges: The Raspberry Pi communicated with the robotic arm over a single-threaded serial port. This limitation made it difficult to run the PID control loop concurrently with other logic, such as returning the arm to a home position. The solution was to create a routine that could rapidly send velocity commands to override the PID data flow when necessary. Other challenges included creating an accurate mapping between the camera’s pixels and the arm’s real-world distance and dealing with hardware failures, such as two servo motors breaking from over-current before software constraints were added.

Project Summary

The “Do Not Disturb” project is a successful proof-of-concept that effectively integrates BCI, deep learning, and robotics to solve a nuanced real-world problem. The development process provided a significant learning experience in complex system integration, PID control algorithms, hardware interfacing with a Raspberry Pi, and robust debugging. The project culminated in winning first place in the “Brain Interaction” robotics category at a competition. The complete codebase, including testing scripts and error handling, is available on GitHub for further review.