The Buzz Kill

Inspiration

The current cold winter makes us miss camping, which inspired us to create a project that honors our love for the outdoors while addressing one of its biggest concerns: safety. Camping is an incredible experience, but it can also feel vulnerable, especially at night. As an all-girls team, safety was especially important to us. We wanted to design a solution that would allow campers to sleep peacefully, while still feeling protected from potential threats like animals or intruders. Our goal was to create something discreet, reliable, and effective. So we created “The Buzz Kill”.

What it does

Our project monitors a campsite while a camper is sleeping in a tent. If an intruder or animal walks past, the system detects both motion and vibration to determine how serious the situation is. A PIR motion sensor detects movement from nearby objects, while an MPU6050 accelerometer measures the intensity of vibrations. This data is classified into high, low or no danger depending on the values. The high danger data is further analyzed using an AI model trained with Edge Impulse, which classifies the situation as low, medium, or high priority. The severity level is then transmitted through Serial communication to a Titan Haptics development kit powered by an ESP32 microcontroller. The Drake HF motor responds by producing short vibration pulses with varying intensity and frequency depending on the danger level. Instead of triggering a loud alarm that could escalate the situation, our system alerts the camper through subtle haptic feedback, ensuring privacy and safety.

How we built it

We built our system using an Arduino Uno Q to manage motion detection and collect sensor data, along with a Titan Haptics development kit featuring the Drake HF motor and ESP32 microcontroller to deliver haptic alerts. The PIR motion sensor detects movement, and the MPU6050 accelerometer measures vibration intensity. We classified danger into three levels: high, medium, and low. High danger was defined as constant motion combined with acceleration over 10 m/s², a threshold we selected after analyzing vibration patterns from simulated threats compared to background noise. Medium priority indicated high acceleration with non-constant motion, suggesting movement nearby but not directly approaching. Low priority is indicated by non-constant motion with medium acceleration, showing movement by a smaller animal. None included negligible acceleration or no motion, accounting for environmental noise since the accelerometer is extremely sensitive. We calculated constant motion by tracking PIR state changes over five 200-millisecond intervals—four state changes indicated consistent movement. We collected our own dataset by simulating different scenarios and exporting the data into CSV files to train our machine learning model in Edge Impulse. Because the Titan Haptics kit operates on ESP32, we used two microcontrollers: the Arduino for processing sensor data and the ESP32 for transmitting the danger signals as haptic feedback.

Challenges we ran into

One of our first challenges was discovering that the Arduino Uno Q does not run on the traditional Arduino IDE, so we had to learn how to use Arduino App Lab, which only worked on one team member’s laptop. The accelerometer’s extreme sensitivity made it difficult to filter out small, irrelevant vibrations, requiring careful calibration and threshold tuning. Integrating two microcontrollers added complexity, especially when establishing reliable Serial communication between them. Additionally, collecting enough high-quality data points to train our machine learning model took significant time and testing. Our AI model to classify different levels of danger was 71% accurate, as we were not able to obtain enough data points without our timeframe to get to higher accuracy. It also functions on a similar model than initially intended, as it classifies data solely based on acceleration, rather than both sensors as initially planned. Furthermore, while integrating our AI model, we were unable to connect the data to the board due to difficulties connecting via SSH. Thus, it is currently being used as a separate model to be integrated.

Accomplishments that we're proud of

We are proud that our system works end-to-end and successfully detects motion and vibration from nearby objects. Our machine learning model aims to accurately classify danger levels based on real-world data that we collected ourselves. We built a fully integrated hardware and AI system from scratch and created a private, non-audible alert mechanism that enhances camper safety.

What we learned

Through this project, we learned how to use the Titan Haptics development kit and Drake HF motor, debug discrete circuits, and manage highly sensitive sensor data. We gained experience training and deploying AI models using Edge Impulse and learned the importance of collecting clean, well-labeled datasets. We also developed skills in managing communication between multiple microcontrollers, and debugging both hardware circuits and software logic pushed us to think critically and problem-solve efficiently.

What's next for The Buzz Kill

In the future, we plan to incorporate Bluetooth communication between the Arduino and the Titan Haptics kit to make the system more seamless and portable. We also want to improve our machine learning model’s accuracy by increasing the sampling frequency and expanding our dataset to include more environmental conditions. Additionally, we aim to refine our accelerometer calibration for more precise vibration measurements and eventually miniaturize the system for practical, real-world camping use