Inspiration
At 2am on the first day of the hackathon, my team determined our project was not viable and withdrew from the hackathon. I decided to continue on solo. I thought, "hey, this isn't the first time I've 'flown solo,' and actually, how can I make this experience better?"
What it does
SaferSolo is a voice assistant system designed specifically for solo pilot operations. SaferSolo can run checklists and remind you if you have forgotten a procedure (many accidents are caused by this). SaferSolo can also remind you what air traffic controllers assigned you to avoid accidental pilot deviations (also a big accident causer), keep track of other traffic around you to increase situation awareness, and assist in emergencies (like a real, reliable copilot!).
How I built it
I use a Raspberry Pi as a lightweight, mobile, computing platform that can integrate with any aircraft communications system. The script is written in Python and takes advantage of Houndify for it's excellent, intelligent speech recognition system. A simple, non-invasive, hardware interface along with both visual and aural feedback makes SaferSolo easy and intuitive to use.
Challenges I ran into
Hotword detection was very difficult to implement as an offline network for recognizing a custom hotword would take a long time to develop within a 36 hour period. After spending several hours testing a wide variety of implementations, I decided to implement hotword detection in a future version. Additionally, allowing for natural phraseology while extracting the particular aviation contexts was quite challenging. Detecting certain keywords while not getting distracted by other filler words required extensive testing. Aviation itself is very regulated in terms of required phraseology, but a real copilot should be spoken to like a normal human. This makes interaction with SaferSolo easier and more intuitive.
Accomplishments that I'm proud of
Checklist mechanics will prevent a pilot from skipping checklist steps or entire checklists. Copilot remembers what phase of flight you are in and will even prompt you of the next checklist to run.
Copilot will listen to air traffic control commands, extract the important information, and remind you what your assigned headings, navigation, or altitude is to prevent pilot deviations from ATC assignments.
You can speak to copilot quickly, fluidly, and naturally. Through the Houndify system, intelligent speech detection is possible with the copoilot allowing for a better and intuitive user experience.
What I learned
Learned how to program python on Raspberry pi (have never done so until Saturday morning) Learned how to implement speech recognition systems Learned theory behind natural language processing
What's next for SaferSolo - Virtual Voice Directed Copilot
Hotword detection, better independent sound card for cockpit integration, faster speech recognition, greater range of features for copilot, more natural speaking voice, avionics integration.
Built With
- houndify
- python
- raspberry-pi
- speech-recognition
- texttospeech
Log in or sign up for Devpost to join the conversation.