The challenge
The client, a North American automotive insurance provider, needed to train their machine learning (ML) model with data representing various, real-world driving actions. Recognizing TELUS Digital’s extensive data collection experience, and impressed by our innovative project plan, the client knew our team of experts would be well suited to meet their objectives. The project included collecting driving data that ranged from mundane scenarios, such as performing right-hand turns at stoplights, to riskier behaviors, such as aggressive lane changes, rapid acceleration and driving while distracted. Several legal and insurance hurdles had to be overcome in order to balance safety and compliance.
The TELUS Digital solution
For this complex project, we deployed a two-pronged approach. First, we conducted over 800 driving sessions in remote, non-controlled environments to track tasks on a variety of roads across 17 U.S. states. The drivers represented a broad range of demographics, drove in varying climate conditions and operated a variety of vehicles to provide adequate data diversity. Second, we hired highly-skilled drivers to perform riskier tasks in a controlled environment (a racetrack) where there was no risk of coming into contact with other drivers. A project manager was onsite to oversee all tasks, and each driver was accompanied by an in-cabin moderator who ensured the directions were understood and followed.
The in-field data collection project included:
- Producing detailed GPS driving reports
- Conducting driver surveys to verify trip duration and ensure compliance
- Employing custom driving prompt instruction modules built to guide participants through their tasks
- Creating custom training materials for trainers, moderators and other team members to guide all facets of the project
- Initiating frequent touchpoints with the client to ensure the data’s usability and validity
The results
Overall, the project yielded exceptional results, enabling our client to make significant strides in the development of their safe-driving application. We anticipate future data collection phases to help further train their model.
- Test drives completed: 2,360 (including long and short trips)
- Events tracked within the driving sessions: 3,800
- Collection duration: 3 months
- Dedicated in-field location (secure racetrack): 1