Autonomous Vehicles

This is quite an interesting field that I want to get into. Combination of AI, software and robotics. Also see Autonomous Racing.

I am breaking into this field through my school’s self-driving car design team: WATonomous.

Resources

• https://github.com/qiaoxu123/Self-Driving-Cars
• Self-Driving Cars seminar by Lex Friedman (2019)
• Lex Friedman, Companies in the self driving space (2018)
• L5kit looks neat
• Other random repos:
  • https://github.com/ser94mor/self-driving-car-using-ros

Datasets

• Waymo Open Dataset
• Argoverse
• Nuscenes
• KITTI

Learned a lot of this listening to George Hotz.

Challenges of Self-Driving Car

Some of the challenges of building self-driving cars (from Cruise talk)

• Low Latency
• Correct (avoid system level false-negatives)
• Robust

1: Static world, solved by mapping 2: Dynamic world, where there are cars around 3: These cars around are reacting to you, potentially solved by reinforcement learning

The self-driving car problem can be divided into 3 components:

• Perception
  • Mapping
  • Localization
  • Object Detection
• Planning (Cars)
  • Prediction (Self-Driving Cars)
  • Behavior
  • Trajectory
• Control
  • Path
  • Velocity

The levels of Self-Driving Car

The Society of Automotive Engineers (SAE) defines 6 levels of driving automation ranging from 0 (fully manual) to 5 (fully autonomous). These levels have been adopted by the U.S. Department of Transportation.

Big companies are moving to combine Perception and Planning into end-to-end control, see Modular vs. End-to-End