Mathias Johanson, Jonas Jalminger Boel Nelson, Tomas Olovsson Joint Subjective and Objective Data Capture and Analytics for Automotive Applications Mathias Johanson, Jonas Jalminger Alkit Communications Mats Gjertz Volvo Car Corporation Emmanuel Frécon RISE SICS Boel Nelson, Tomas Olovsson Chalmers University of Technology
Objectives How can we collect high-quality subjective user experience data and driving condition data? (Traditionally done by interviews and clinics.) How can we scale up this data collection and make data quality higher? (Capture user experience data closer to the actual use of the product.) How can we analyze subjective and objective data together to increase knowledge about how products are used and experienced? How can we improve Active Safety and AD systems (and thereby traffic safety) based on feedback of user experience data and measurement data? How can we shorten development cycles by continuous improvements of software, supported by connectivity and telematics services? How can we preserve the privacy of users while capturing subjective and objective data?
Approach Montrig Telematics unit Engineer Approach Measurement assignment design Survey design Analytics design Back-end server infrastructure Montrig Subjective / objective data analyses Analytics fw (triggering subjective user feedback) Monitored signals Answers uploaded Questions sent to app Bulk measurement data upload Was the alert relevant ? Yes No In-vehicle signals monitored and logged Telematics unit Driver’s smartphone with subjective data capture app Test vehicle
Concept Need for new knowledge identified How do customers experience our products? How are the vehicular subsystems performing? Concept WICE telematics & remote software download Need for new knowledge identified Design subjective & objective data capture tasks Measure subjective and objective data Analyze subjective & objective data Improve vehicular software based on analyses and ML Roll out new sw versions to test vehicles Data capture configuration and survey design tools WICE in-vehicle data logging & telematics Smartphone app Rapid prototyping framework External data sources Cloud-based analytics framework and methodology ML training data sets
Framework Architecture Users Presentation layer / User interface Back-end server architecture Measurement Task design tool Poll design tool Monitoring & Triggering mechanism Measurement Task Manager Subjective Data Task Manager Analytics Framework Telematic service layer Smartphone App Service Layer ? ? ? Data sources
Smartphone App Development Hybrid apps Multi-platforms: iOS Android Windows Phone Web View Ionic (Platform look & feel) Angular (Clean MVC design) Cordova (Hardware abstraction) Bluetooth Camera for barcode scanning GPS Notifications Server Backend Swagger (API Dev) Java
Register and manage cars
Menu and Profile
Polls when?
Poll Question Types yes / no multiple choice rating
Pilot Use Cases Two focused active safety use cases: Driver Alert (DAC) and Forward Collision Warning (FCW)
Joint Subjective / Objective Data Analytics Current analysis framework can produce (2D & 3D) histograms, time plots, contour plots, heat maps Subjective and objective data can be analyzed together Answers vs. User info Answers vs. In-vehicle (time series) signals
Privacy and integrity issues When subjective data capture is scaled up to large customer groups, privacy issues must be considered Approach is to use differential privacy
Revised Framework Architecture Users Presentation layer / User interface Back-end server architecture Measurement Task design tool Poll design tool Measurement Task Manager Subjective Data Task Manager Analytics Framework Privacy preservation layer Privacy preservation layer Telematic service layer Monitoring & Triggering mechanism Smartphone App Service Layer ? ? ? Data sources
Conclusions Access to high quality data (both subjective & objective) improves vehicle development (cf. knowledge-driven product development) Proof-of-concept implementation shows that subjective and objective data can be captured and analyzed togehter to improve data quality Supports Rapid Prototyping of new Active Safety (and other) functions System can be used to capture training data sets for Machine Learning algorithms in Active Safety and AD systems Supports Continuous Deployment of software in test vehicle fleets Improved connected active safety and AD systems improves traffic safety Contributes heavily to digitization, leveraging IoT, ML and Cloud Computing technology for vehicular applications
Thank you!