Case Study

Field Operation Testing (FOT)

Project Background:

  • This major Japanese Automotive OEM has been developing a state-of-the-art Advanced Driver Assistance System (ADAS) system for their new model year vehicles.
  • To train the autonomous algorithms in the vehicle, extensive driving and mileage accumulation was needed in multiple global regions.
  • As a trusted partner, SCAE was selected to manage the global ed-to-end execution in Japan, China, Europe, and the US.

Customer Goals:

  • Collect video data of driving activities in each region, on a continuous basis, for events/environments of interest for system training
  • Utilize activity output data (annotated video frame images) at development HQ

Details of the work done by SCAEUS:

  • End-to-end service for Route-planning, Vehicle Prototyping, Test
    • Drive, Data Management & Analysis
      Plan proper routes to capture the events and edge-cases of interest for system training
    • Establish the standards and operating manuals to conduct testing as desired by the client.
    • Retrofit production vehicles with next-gen production intent sensors and hardware (5 total vehicles for the US region)
    • Secure drivers and resources to execute driving activities, complying with local laws and regulations, e.g. CCPA, PIPL, GDPR (data anonymization and data sanitation), and permit & approvals to operate uncertified vehicles on public roads.
    • Collect and upload video files / driver reports indicating any anomalous vehicle behavior or events.
      Annotation of video fames for system training
  • Communication and reporting in multiple languages (English/Japanese), leading global meetings with client.

Outcome & Customer Results:

  • Prototyping/Retrofit – 5 vehicles retrofit with next-gen sensors, ready for on-road testing.
  • Route Planning – established key routes in multiple regions to cover 8 months of data collection, which achieves desired events/edge cases.
  • Data Collection – Collected videos from vehicles totaling >60TB daily.
  • Data Handing – Designed and architected the data collection/handling methods and SSD logistics with upload station to optimize SSD counts and CAPEX requirements.
  • Data Sanitation – Anonymization of video frames for compliance to
  • Data Annotation – Post-process images for up to 60 classes of objects using 3D bounded box and semantic segmentation methods.