Skip navigation
AI for Full-Self Driving
Video

AI for Full-Self Driving

Matroid, 2020

Read offline

auto-generated audio
auto-generated audio

Editorial Rating

8

Qualities

  • Innovative
  • Scientific
  • Insider's Take

Recommendation

Tesla is at the leading edge of making self-driving cars a reality. By introducing Smart Summon in 2019, the company deployed the first real-world self-driving technology available to consumers. Through overnight updates, new elements are deployed regularly. While many car makers are increasing the range and sophistication of sensors to solve complex challenges, Tesla is focusing on a largely visual approach and discovering that a picture is worth more than a thousand lines of code. Though dense and fast-paced, this talk by Tesla’s AI leader Andrej Karpathy illuminates for the non-specialist some of the technical challenges Tesla is confronting as it develops viable self-driving cars.

Summary

Tesla autopilot research relies on a visual approach to achieving self-driving capability.

Tesla’s research into car autopilots is based on a fundamentally different technology than other manufacturers of self-driving vehicles.

The use of high definition maps and LIDAR, where laser beams map the local obstacles, is common. By contrast, Tesla is focusing on a visual-only approach. In this approach, the data from cameras is the main basis for navigating locally. Low definition maps aid wider navigation.

Detectors are designed by using Tesla’s currently active fleet to gather data on the success or failure of algorithms.

...

About the Speaker

Andrej Karpathy is a specialist in deep learning. He is senior director of AI at Tesla and leads the computer vision team of Tesla Autopilot.


Comment on this summary