Autonomous vehicles need to be safe in any situation they encounter. Simulation plays a vital role in ensuring test coverage of all scenarios. Simulating traffic isn’t just about rendering cars and roads — it’s about controlling how every actor moves through a scene. At Uber ATG, setting up these motions could take anywhere from 30 minutes to 16 hours, making it nearly impossible to plan simulation work at scale.
For this project, I conducted research and testing to identify major contributors to simulation creation time. The solution incorporated inputs from evolving needs of operations teams, vehicle dynamics and data requirements from systems engineers, and best practices from experts in simulation and game-industry tooling. The resulting tools made simulation creation faster, while also making actor motion setup more predictable and far less frustrating for users.
Product owner responsible for research of evolving user process and needs, defining feature behavior through mockups and PRDs, and refining backlogs with developer leadership.
1 product owner and 10 simulation developers.
Additional 1 product designer and 1 user researcher for last month of project.
2 months for research, prototype feedback and iteration, and PRD approval.
1 month for UI design and testing.
Decreased maximum simulation-creation time by 80%. Also unblocking the capability to simulation 5 new scenario types, translating to hundreds of additional tests.