Autonomous Driving Tools

I spent two years as an engineering manager at Level 5, Lyft's (and later Toyota's), self-driving car division. I lead one of the data viz and tools teams that AV researchers used to perform their research.

The tools we built were used all over the company to visualize what the autonomous vehicle was doing in realtime, as well as replay data from previous runs. Eventually our tool grew to encompass searching across multiple runs and replaying gigabytes of data streamed from our AWS buckets. When that was too slow for engineers trapped at home during Covid, we pioneered a hybrid local react + VNC-like streaming video solution from cloud based workstations. This let our users interact with massive datasets from any device, even an iPad.

Co-Pilot

The Co-Pilot view shows what the AV is doing in realtime. It was used by developers in the vehicle during test runs, as well as reviewing the recorded runs later.

CarFlix Mission Replay

Playfully named CarFlix, the Mission Replay tool let researchers search through terabytes of sensor and camera data across multiple runs (~1GB/minute of real world drive time)

Sim Editor

Sim Editor was a tool to edit maps and scenarios for simulated av runs that were used to train the driving models.

AV Studio

AV Studio was our final tool during my time at Lyft. It let researcher customize data visualization for the particular vehicle being tested, since we knew we would have many models under use inside of Toyota.