RoadBuddy

A buddy for everyone that will keep you alert during your trip without them nagging you about how bad of a driver you are!

Comment

A problem we have personally experienced ...

Accidents due to drowsy drivers continue to increase yearly. Those are accounted for 72,000 in 2013, and 91,000 in 2017. It is more than 800 fatalities each year! - NHTSA

We know it is dangerous - but we still do it anyway. Approximately 50% of US drivers admit to driving while being drowsy. 20% admitted to falling asleep behind the wheel at some point in the past year. Drowsy driving affects everyone, but particularly those driving alone between midnight and 6 a.m., or in the late afternoon.

Core concept behind our RoadBuddy!

Our solution: A buddy for everyone that will keep you alert during your trip without them nagging you about how bad of a driver you are!`

We all know that prevention is more efficient than cure, we aimed to develop a pattern of active prevention. Our idea is to do timely check-ins to make sure you are all-right, and suggestions to take a break from driving. This happens every hour of driving, and can easily be modified through the user personal settings.

Aside from active prevention, we also looked into a way to monitor the driver directly. The idea of drowsiness detection through facial recognition is pretty widespread technically speaking, but very unlikely to be accepted by a community caring about their privacy. We thus oriented our approach to the previous work of one of our team-member: Using the heart rate variability signal to monitor the driver's condition. The information about BPM (R-R peaks distances) is enough to determine a drowsy state, as we all tend to physiologically have a steadily decreasing rate when fatigue catches up. For your own interest, here is a list of papers supporting this idea:

On our end, we chose to implement this approach through our connected watches (Fitbit in this case), but that would also be feasible with an Apple Watch or any similar smartwatch.

Where the fun is!

  • PLAY A GAME OR ASK FOR JOKES AND RIDDLES: Keep yourself busy by playing games like 'Name That Tune', or joke away with RoadBuddy.

  • LEARN SOMETHING NEW: Hearing fun facts, or something interesting about the area you are driving in will surely keep you intrigued and alert.

  • CALL A FRIEND (OR FAMILY): RoadBuddy can intelligently suggest someone you could call, and propose interesting discussion topics.

The technology behind it ...

Because of hands-free settings, it is all handled by voice. So the first step was to have a functional app threading both speech-to-text and text-to-speech technology. Once done, we hosted all the possible content generation through AWS lambda functions (and API gateways) to easily communicate with the Android app. The text is analyzed (currently the good old-fashioned machine-learning of if-else statements, but totally could be improved) to discuss with the user about what its preferences are. We can easily generate and stream a joke or a fun fact, and use the Smartcar API to get access to both the car charge (that we also use as trigger in a message for a break) and the location (that we plug into a HERE API giving us the places around) to give short descriptions of the surroundings (thanks to a Wikipedia API). As game, we also implemented a blind test through a YouTube streamer.

Challenges we ran into

We faced challenges authenticating to API and make our conversation flow smooth. Also, deciding to deploy everything in the cloud brought some difficulties as we had to learn some of the AWS services from scratch.

Accomplishments that we're proud of

Definitely our idea and the fact that everything runs in production. Also, our chat agent has a few good jokes that helped us stay in the run!

What we learned

We learned how to manage our own conversation flow. Use machine learning in the area of driving voice control.

What's next for RoadBuddy

We would definitely continue to work on it as RoadBuddy really drove us to make solo traveler more attentive and road safer. That is really what we aimed for.

Built With

Share this project:

Updates