CaveMan

The Mission

• Map out unexplored Caves
• Search and Rescue mission

The Challenges

• Locating people in a dark environment with potential rockfall
• 5G connectivity issues in the cave

The Approach

• Swarm of drones, some router drones some detection drones.
• The drones chain into the cave to keep signal strength to leverage 5G edge computing via the router drones for AI models
• To overcome visibility issues we navigate and explore based on sound instead.
• Use directional microphones on drone in conjunction with AI Model to control drone.
• The model is trained to detect Human sounds.

The Architecture

• Use CNN on spectrogram of audio data.
• The CNN is able to use spectrograms to detect human sounds from ambient background noise and echoes in caves.
• The AI model will also used to give range predictions in order to distinguish loud shouting that might be far away compared to quitter speech which may be closer which would make amplitude alone not a good enough measure of distance.
• Directional microphones can then use the closest distance estimate for control

Next Steps

• Test models in simulation to test the direction microphone system.
• Collect and create more realistic data to train models on.
• Develop Network software system.

Built With

• javascript
• python
• pytorch
• sklearn