22
Autonomous Driving and HD Maps Liang Heng Co-Founder & CTO RoadStar.AI

Autonomous Driving and HD Maps

  • Upload
    others

  • View
    5

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Autonomous Driving and HD Maps

Autonomous Driving and HD Maps

Liang Heng

Co-Founder & CTO

RoadStar.AI

Page 2: Autonomous Driving and HD Maps

Why Autonomous Driving?

• Fewer accidents

• Road crashes lead to 3000 deaths and 100,000 injured worldwide a day.

• Decreased traffic congestion

• Enhanced human productivity

• Reduced cost of ride sharing

• Fewer parking lots, and no more parking headaches

• Improved mobility for the people who cannot drive

Page 3: Autonomous Driving and HD Maps

History of Autonomous Driving1986

CMU Navlab 12005

Stanford Stanley1st in DARPA Grand Challenge

2007CMU Boss

1st in DARPA Urban Challenge

Page 4: Autonomous Driving and HD Maps

Different Levels of Autonomy

Page 5: Autonomous Driving and HD Maps

Two Different Approaches

Page 6: Autonomous Driving and HD Maps

Level 4 Players in the US

Google/Waymo

Zoox Cruise

Uber

Page 7: Autonomous Driving and HD Maps

Comparison of Level 4 Players in the US

Page 8: Autonomous Driving and HD Maps

Level 4 Players in ChinaBaidu

Commercialization in 3 years,mass production in 5 years.

UiSeeUrban Mobile Box

iDriver+Autonomous Street Sweeper

Page 9: Autonomous Driving and HD Maps

Industry Map

• a

Page 10: Autonomous Driving and HD Maps

How do Autonomous Cars Work?

HD Maps

Sensing

Perception

Localization

Planning Control

Page 11: Autonomous Driving and HD Maps

Sensors on an Autonomous Car

Page 12: Autonomous Driving and HD Maps

Perception

• Dynamic objects• Detection

• Classification

• Tracking

• Predication

• Traffic sign & light recognition

Page 13: Autonomous Driving and HD Maps

Localization: Differential GNSS

Page 14: Autonomous Driving and HD Maps

Localization: LiDAR Reflectivity Map Matching

LiDAR reflectivity map Correlation of real-time scan and map

Page 15: Autonomous Driving and HD Maps

Mission & Path Planning

• Traffic rules / logics

• Obstacle avoidance

• Safety metrics

• Vehicle dynamics

Page 16: Autonomous Driving and HD Maps

Control

• Vehicle dynamic models

• Control algorithms• Traditional PID control

• Dynamic perfect control

• Neural network control

Control Algorithm

Position

Orientation

Velocity

Inputs

Throttle

Brake

Steering

Gear

Outputs

CAN bus

Page 17: Autonomous Driving and HD Maps

High-Definition (HD) Maps

• Road network layer• Traffic sign, lights, poles

• Lane markers

• Logics

• Perception layer• Static features

• Grounds

• Localization layer• Reflectivity

• Geometry

Page 18: Autonomous Driving and HD Maps

HD Map Generation Pipeline

Data collection3D

reconstruction Feature

extractionLabeling

Road network layer

Reflectivitystatistics

Ground detection

Perception layer

Localization layer

Page 19: Autonomous Driving and HD Maps

Efficient 3D Representation Using Octree

Page 20: Autonomous Driving and HD Maps

Road Network Representation

Page 21: Autonomous Driving and HD Maps

Challenges and Potential Improvements

• Construction zones:• Additional radio signals to guide traffic, and

• Construction plans filed in advance and available to autonomous cars.

• Traffic lights:• Radio signals to indicate state, or at least

• Clear visibility to car’s cameras, regardless of the sun’s position.

• School bus, ambulance, and other emergency service vehicles:• Additional radio signals to indicate state.