Jett-Sen
2020
Enabling a mobile sensory system
Jett-Sen is a sensor platform developed for Panasonic’s electrically assisted bikes. The goal of the platform is to turn electrically assisted / shared bikes into an instrument that is capable of understanding user driving patterns, city conditions and city dynamics in order to enable smarter, unbiased, resilient and logic driven mobility systems and urban interventions.
The Continuous Data Collection System Design
What sensors does the system have?
| Bike State
Battery Status
Mechanical Status
Electrical Status
Driving Mode Status
Headlight Status
| | Environmental State
3 Axis Accelerometer
Temperature
Light
Humidity
Pressure
| | | Street / Geographical State
Front Facing Camera
GPS Unit
System Architecture Diagram
Development of Unbiased Driver Assistance and Driving Pattern Prediction
The second phase of the project aims to develop an unbiased driver assistance system. To do so, we feed the data from the bike directly into an unsupervised learning model (K means) which helps classify different bike states throughout a single bike ride.
Once classes are created in an unsupervised manner, they are fed into a Recurrent Neural Network for them to be used for sequential bike trip prediction. In this way, the RNN model classifies with the classes created in an unsupervised manner.
Model uses unsupervised learning to create clusters / classes that are then fed into an RNN that is capable of predicting classes in a sequenced manner.
Classification Results
Plotted clusters generated by K means algorithm. Empirical observation proves that classes make sense in the context of a bike trip.
Recurrent Neural Network Prediction Results and Training
Updates to the project will be coming soon…
Credits and Acknowledgements
Yasushi Sakai (Project Vision & Implementation) | Nanako Yamazaki (Project Vision and Panasonic Enabler)