Rocko | Devpost

Inspiration

Amr fell asleep to a video about the Nutty Putty Caving Disaster. He pitched the idea and we were all intrigued.

What it does

Rocko is an emergency beacon that sends an AI-classified distress message through solid rock by encoding your spoken emergency into two bytes and transmitting them as a magnetic signal to a surface receiver. It works by sending the diagnosis instead of the raw data, running on a QNX OS for reliability.

How we built it

We used 20 meters of 18-gauge insulated copper wire coiled around a CD storage cylinder for the magnetic transmitter, and a QNX equipped Raspberry Pi 5 brain. Our speech to text uses OpenAI Whisper, and our injury classification is done through a locally built and trained MobileNetV2 convolutional neural network with a single trained softmax layer on top. The input is a 224x224 wound photo and the output is a predicted class of injury, exported to TFLite to run on the Pi. The encoding schema uses a tilde to signify the beginning of a message. If there are four zeros after the tilde, this is a heartbeat message. For emergency messages, the four following bits represent the emergency type (lost, trapped, injury type, etc.). The signal is sent at 8Hz, and uses Manchester encoding of the 12-bit message (tilde included) to ensure distinguishability of each bit. The receiver is a quantum TMR magnetometer with a bandpass filter between 7-9Hz to filter noise.

Challenges we ran into

Anything you could possibly think of (Camera module not functioning, Raspberry Pi configuration issues, transmitter coil not working, parts unavailable, etc.). All these challenges were overcome thankfully due to the hard work and resourcefulness of our team.

Accomplishments that we're proud of

Anything you could possibly think of (Accurate injury classification, successful transmission and decoding of message from distance through walls, proper speech recognition and emergency identification, two entirely local AI architectures running on Raspberry Pi).

What we learned

 - QNX configuration
 - Message transmission through copper wire coil
 - Manchester encoding
 - Hardware assembly
 - Neural network development
 - Signal processing
 - Matched filtering
 - Mission critical design principles
 - Resource allocation
 - Quantum mechanical phenomena (quantum tunneling)

What's next for Rocko

After receiving the signal, the surface-level receiver should relay the emergency message, along with it's own location to the appropriate EMS, utilizing higher frequency radio waves. Ideally, the surface receiver should turn into multiple surface level nodes to increase coverage area and allow for accurate triangulation to increase rescue speed.