"Smart Cars of Tomorrow: Learning Driving Patterns From Sensor Data" by Ellie Dobson, Data Scientist at Pivotal

There is a wealth of insight to be gained from data collected during drive time. For our prototype, we use cheap Bluetooth dongles together with free Android apps to stream data from the Electric Control Units of our cars. Using Machine Learning, we can derive individual driver ‘fingerprints’ from variables such as speed, acceleration, driving times and location.

In particular, we will demonstrate the models we have built on the recorded data. Our models predict the destination of the journey at its start time, and thus how far it is expected to travel before fuel or electricity runs out.

Transcript

1. Real-Time Journey Prediction From Car-Sensor Data 2014-10-06 Ellie Dobson edobson@pivotal.io
2. MICHAEL  MINELLA   RONERT  OBST   PHIL  BERMAN   DEREK  BEAUREGARD   ALEXANDER  KAGOSHIMA   ELLIE  DOBSON   ALEX  KRAMER   GEOFF  PLEISS   MEREDITH  PRETORIUS   RICK  REILLY   ARIELLE  SILVERMAN   GEORGE  FOSTER   JAMIE  OMEARA   ROBBIE  CLUTTON   VICKIE  FU   DARREL  SHARP   MARK  RUESINK   BRIAN  DUSSAULT   STEVEN  BECKER   AUSTIN  RUTHERFORD   MICHAEL  NATUSCH   PAUL  FARINO   IMTIAZ  JAFFER   CATE  ZOVOD   MEGAN  WHEELER  
4. 01 0D! 01 0D! 18 DA F1 11 03 41 0D 30! >!
5. R E A LT I M E RANGE PREDICTION
6. THE INITIAL IDEA
7. My old commute My commute Alex’s commute
8. BUILDING THE MODEL
10. {! "vehicle_speed":103,! "obd_standards":2,! "intake_manifold_pressure":"",! "accelerator_throttle_pos_e":14,! "engine_load":89,! "maf_airflow":33,! "latitude":"32.897554",! "vin":"1HGCM82633A004352",! "bearing":"343.922580",! "catalyst_temp":779,! "relative_throttle_pos":12,! "fuel_level_input":89,! "fuel_system_status":[2,0],! "accelerator_throttle_pos_d":29,! "acceleration":"0.953",! "throttle_position":21,! "barometric_pressure":97,! "control_module_voltage":13,! "longitude":"-96.810236",! "distance_with_mil_on":0,! "coolant_temp":94,! "intake_air_temp":34,! "rpm":1593,! "short_term_fuel":-2,! "time_since_engine_start":4054,! "absolute_throttle_pos_b":38,! "long_term_fuel":3! }!
11. 1. Labelling the journey cluster
12. 2. Classification model for journey cluster based on start
13. 3. Classification model for journey cluster based on location
14. MARSHALLING THE DATA
15. TCP SOCKET OVER BLUETOOTH HTTP POST OVER CELLULAR
16. HTTP transformers rabbit rabbit file gemfire tap python file system GemFire Spring XD Admin | | |
17. REALTIME PREDICTION
18. 1. Online destination prediction at start
19. 2. Online update of destination prediction
20. 3. Range prediction from journey
21. LOOKING INTO THE FUTURE
22. D R I V I N G EFFICIENCY G A S PRICING T R A F F I C PATTERNS A C C I D E N T ASSISTANCE A D D I T I O N A L USE CASES
23. Ellie Dobson edobson@pivotal.io
24. Model Overview
25. Car Data Collection Data Streaming Data Science Batch Training Real-Time Prediction Dashboard Field Engineering Architecture Backend Data Processing Data Serving Field Engineering