Chapter

Waymo's Approach to Perception and Planning
listen on Spotify
53:46 - 57:44 (03:58)

Waymo's self-driving cars have a perception stack and planner, but the line between the two is not always clear. The company relies on a combination of hand-engineering and machine learning to make its vehicles more autonomous.

Clips
The two types of simulators used in building perception systems are those that work on real data and those that don't work on real data.
53:46 - 55:38 (01:51)
listen on Spotify
Perception Systems
Summary

The two types of simulators used in building perception systems are those that work on real data and those that don't work on real data. By defining the state vector that separates the best team to build a perception system from the best team to build a planner, they can replay their perception data and test their system with different objects in the scene over time.

Chapter
Waymo's Approach to Perception and Planning
Episode
George Hotz: Comma.ai, OpenPilot, and Autonomous Vehicles
Podcast
Lex Fridman Podcast
The output of a perception system cannot be completely expressed in a spec document.
55:38 - 57:44 (02:06)
listen on Spotify
Perception Technology
Summary

The output of a perception system cannot be completely expressed in a spec document. Hand-engineered systems have limitations, especially with occluded areas.

Chapter
Waymo's Approach to Perception and Planning
Episode
George Hotz: Comma.ai, OpenPilot, and Autonomous Vehicles
Podcast
Lex Fridman Podcast