When Musk first heard about Wayve’s AI deep-learning approach to autonomous driving, he wasn’t sold on it. Instead, Tesla opted for a rules-based approach in which separate modules are used for perception, planning, and control of a vehicle.
Wayve, by contrast, employs a self-learning AI system whereby raw data from a vehicle’s sensors is fed into a neural network, and from there, deep-learning models handle the vehicle navigation. The system, also referred to as “end-to-end deep learning,” is designed to respond to real-world complexities the way human drivers do, rather than simply follow rules that might not account for those complexities.
“None of the way the car’s controlled, in terms of the speed it chooses or the lane it takes, none of this is hand programmed,” he said. “It’s not following a map, but it’s making all these decisions based on what it sees.”
At the Brainstorm event Tuesday, Kendall reiterated that sentiment, saying it was “gratifying” that Tesla made its pivot. But Kendall isn’t one to gloat. Autonomous driving, he said, is “one of the greatest engineering problems there is today,” and he expects “more challenges” on the road ahead. With that in mind, “one of the key values we have at our company is humility, and really not assuming we know the solutions.”
