S elf-driving cars must navigate dynamic environments under variable conditions—night and day, summer and winter. To accomplish this task, each car uses multiple devices designed to identify specific object types (i.e., pedestrians, other vehicles, road signs, painted roadway markings, etc.) at specific distances and conditions.
The amount of data generated by each device varies, and of course data will vary by environment—a quiet rural road will generate less data than a busy urban setting. Here are some typical data rates our customers are seeing with various devices.
A typical test car uses about 20 (+/-) devices total and commonly generates between 8-10TB of data daily, with some customers generating more depending on their devices and driving profiles. At these daily data rates, each test car will generate between 2-2.5PB of data in a year, assuming it runs 5 days per week. Add more devices, increase resolutions, add more test vehicles, and we can co-opt the famous quote popularly attributed to Everett Dirksen to read: a petabyte here, a petabyte there, and pretty soon you’re talking about real data.
As the race to deliver the first fully autonomous vehicle heats up and data is the fuel powering it, companies are ramping up their test hours and the number of vehicles in their test fleets—dramatically accelerating data growth.
Quantum’s expertise in the end-to-end management of this critical data (from test vehicle, to engineering test and development, to long-term retention) is proving to be a key enabler for accelerating the autonomous vehicle development process. It’s fun participating in such an exciting and dynamic market, and it’s personally rewarding since my career started as an engineer in the auto industry—way back when ABS brakes were the hot new technology.