Inspiration
We love using the rentable scooters to get around campus. A big fear though is getting hurt. We wanted to create something that could give people the joy of riding on scooters while alleviating the fear of getting hurt.
What it does
It detects if the user crashed and sends out a signal if that is the case. The signal is then used by ElevenLabs to send a message alerting that a crash was detected and asking if the rider is okay.
How we built it
We used the SensorTile Box Pro and its built in accelerometer and gyroscope with an AI model trained on collected crash data to detect when a crash has happened. The model was trained by ST AIOT Craft on the data sets we made.
Challenges we ran into
One of our challenges was collecting enough data for our model to reliably detect when a crash has occurred. This involved uploading multiple data sets for the classes hard stop, left tip, and right tip. In the end our model was able to correctly detect a crash with 80% accuracy.
Accomplishments that we're proud of
We are proud of how quickly we were able to learn how to use the SensorTile Box and train an AI data using ST AIOT Craft. To be able to create and train an AI model to have a high accuracy in 24 hours was an impressive task we were able to accomplish.
What we learned
We learned how difficult it is to detect the difference between a crash and normal operation. To only alert when a crash was detected, we had to train the model on what a normal tip and stop looks like so that it could differentiate that from the tipping and stops caused by a crash.
What's next for MangledPi
The next steps for MangledPi is to implement the Raspberry Pi as a gateway and deploy our project through cloud connection.
Built With
- python
- sensortile
- stm
