Just How Much Data Is Required to Make Autonomous Vehicles Truly Road-Ready?

Abstract: AI models guiding autonomous vehicle systems must be trained to anticipate and react correctly (i.e. safely) when faced with any of the myriad road scenarios they could encounter. Industry leaders are harnessing ever-greater volumes of data in order to perform this training, exposing AI to vast image datasets of known road factors and collected experiences that AI can call upon when interpreting given scenarios.
This exposure to new data is essential to the pursuit of Level 3 (L3) autonomous driving, in which vehicles are able to take responsibility for all safety-critical functions under certain conditions (but where the human driver must take over when the AI encounters an unfamiliar situation). AI is already able to solve many discrete driving issues, such as visually recognizing traffic light colors and obeying those rules. However, the tougher challenge is that, while image dataset volumes are exponentially increasing, this growth leads to diminishing returns as far as how much that data actually improves an AI model. And while L3 autonomous driving means that AI can navigate safely in all but less common “edge cases,” the reality is that driving in the real world means encountering such edge cases fairly frequently.
Because of this, autonomous vehicle companies face significant questions as to how to know when their AI has reached the threshold of being safe enough for the road – and how much data that is going to take. This question is exacerbated by ongoing consumer expectations around (and apprehension about) self-driving cars, where even achieving accident rates that are twice or three times safer than human drivers may still not be enough to win their trust.
The presentation will address the question of how much data will be required to succeed in reaching L3, and possible paths to get there and beyond.

Bio: Alexandr Wang is the CEO and a co-founder of Scale, which accelerates its customers' AI development by democratizing access to intelligent data (companies using Scale include Lyft, Cruise, and nuTonomy). After dropping out of M.I.T. to become a teenage tech lead at Quora, Alexandr founded Scale in 2016 and became the youngest entrepreneur ever funded by Accel. Named a Forbes 30 Under 30 honoree, Alexandr is a frequent speaker on the current AI challenges faced by autonomous vehicle innovators. He lives in San Francisco, where Scale is headquartered.

Open Data Science Conference