Autonomous cars

Šimon Rozsíval

Outline

  • What is a self driving car
  • History
  • How does an autonomous car work
  • Legal and ethical questions

Levels of Autonomous Vehicles

  • National Highway Traffic Safety Administration - 2013
  • Level 0 - No-Automation
  • Level 1 – Function-specific Automation
    • Parking assistant, cruise control
  • Level 2 - Combined Function Automation
    • Adaptive cruise control, lane centering
  • Level 3 - Limited Self-Driving Automation
    • No driver needed most of the time
  • Level 4 - Full Self-Driving Automation
    • No driver needed at all times

First cars

  • 1769 - the first steam-powered automobile
  • 1807 - internal combustion engine - hydrogen
  • 1864 - gasoline powered combustion engine
  • 1886 - first "production" car - several identical copies
  • 1888 - first electric car

Automation

  • 1921 - invention of automatic transmission
  • 1940 - first cars with automatic transmission
  • 1986 - Navlab 1 - Carnegie Mellon University
    • ​36 kph

Automation

  • 1921 - invention of automatic transmission
  • 1940 - first cars with automatic transmission
  • 1986 - Navlab 1 - Carnegie Mellon University
    • ​36 kph
    • ALVINN
    • D*

Grand Challenge

  • DARPA
  • $1 Million prize
  • third of military vehicles autonomous by 2015
  • 2004 - nobody finished
  • 2005 - all but one finished
  • 2007 - urban challenge
    • avg speed of the #1: 22.53 kph

Google Self-Driving Car Project

  • Google works on creating fully autonomous car
  • 1.5M miles driven
  • car designed from
    scratch
    • sensors first

Other automakers

  • Tesla Motors
    • Fleet learning => ~100M miles in autonomous mode
  • Audi A7 "Jack" - level 3
  • Volvo XC90
  • BMW iNEXT - 2021

Czech Republic

  • Škoda
    • line centering, other assistants
  • Valeo
  • RoboAuto
    • academic research
    • VUT Brno + Artin
  • FEL ČVUT

What it takes to build an autonomous car

Path planning

  • How to get from point A to point B
  • Easy task - many navigation programs in use every day bi millions of drivers

Precision locating

  • locate on the map
  • position on the road
    • current line
  • distance from other traffic participants

Maneuvering

  • steering the wheels
  • pushing the pedals
  • changing driving mode
    • drive
    • reverse
    • park
  • turning on/off different types of lights
    • signaling change of direction
    • being visible
  • how long to perform each action
  • how to combine the actions

Adapting to real world

  • even if you plan to go in a certain direction, you are not guaranteed to go there
    • malfunction
    • incorrect measurements from sensors
    • external conditions
  • how does the car behave when one of the brakes does not work?

Be allowed to drive

  • you cannot drive your AV everywhere
    • even for testing
  • bureaucracy
  • the systems must work well and be durable

Persuade people to use the car

  • not everyone trusts AV
    • "75% of US drivers"
  • "When my personal computer does not work well all the time, how shout I expect the AV to be any different?"

How does the car see the world arround?

Radar

Lidar

Camera

  • detection of traffic signs and traffic lights
  • detection of non-moving obstacles
  • stereo cameras - distance measurement
  • what spectrum does it operate in?

Pattern recognition

  • Traffic Sign Recognition (TSR)
    • find and classify traffic signs in a photo
  • Identify objects based on the partial 3D shape

Environment modeling

  • other cars
  • pedestrians
  • cyclists
  • crossroads
  • traffic lights
  • street signs
  • children & animals
  • railroad crossings
  • construction works
  • local driving rules
  • emergency vehicles
  • driving schools

Odometry

Odometry is the use of data from motion sensors to estimate change in position over time.

Predicting traffic

  • Avoid collision paths
  • Avoid blind spots
    • The humans do not have 360° view and use mirrors instead
  • Goals of other traffic participants
    • Anticipate trajectories of others
  • Predict sudden changes of direction
    • Prospect Theory - taking risks

Driving strategies

Defensive driving

  • Avoid blind spot
  • Keep distance from other cars
  • Anticipate sudden changes of direction
  • Pedestrians or animals entering the road
    • children near playgrounds
    • animals in a forest at night

Damage control

  • stop to avoid secondary collisions
  • break to minimize impact velocity
  • steer the car to collide 
    • sides are most vulnerable

Feel natural to the passenger

  • The passenger must feel comfortable
  • Be predictable
  • Try to drive like a human would

Talk to others

  • Automated cars can talk to each other
    • Cars can know exactly where others will go and not only predict their paths
    • Coordinated 
  • Sharing environment data
    • car can see further
    • not having to stop at the crossroads

Law and Ethics

Autonomous cars

on public roads

  • Nevada (2011)
    • testing on public roads - Google
  • United Kingdom (2013)
  • France (2014)
  • California
    • completely driverless cars in 2018 (not yet passed)

Can you make the car break the rules?

  • Speeding
  • Parking at forbidden places
  • Going past "No entry" sign
  • "Do not use EET" button on new Czech cash registers...

Loss of jobs

  • Taxi drivers
  • Truck drivers
  • Bus drivers
  • Not a new (ethical) problem
    • discussion about basic income
    • "electronic persons" - should pay social security and health insurance

Increasing mileage

  • "Send your car to go shopping and then find itself a parking spot where it would wait until you need it."
  • More traffic
  • More pollution

Who would be responsible for a car accident?

  • The driver (who was asleep at the moment)?
  • The automaker?
    • manufacturers can already be held responsible for safety defects in the cars - e. g. Ford Pinto in 
  • The human driver in the other car?

The trolley problem

  • philosopher Philippa Foot - 1967
  • several scenarios
    • two tracks
    • fat man
    • fat villain
    • looping

Objections to the TP

  • Even humans cannot choose the right answer
  • The AI does not have the full information at the moment of taking the decision
    • age, gender, number of children, ...
  • There is no correct answer => the AI cannot chose the right answer
  • AI does not have to make the correct decision, but rather not the wrong decision
  • Low probability of such situation

Questions

Sources

  • Google Self driving car
    • https://www.google.com/selfdrivingcar
  • Autonomous Vehicle Implementation Predictions
    • Todd Litman, Victoria Transport Policy Institute
    • http://www.vtpi.org/avip.pdf
  • National Highway Traffic Safety Administration
    • http://www.nhtsa.gov/Research/Crash-Avoidance/Automated-Vehicles

Sources

  • The Simple Solution to Traffic
    • CP Grey - Youtube channel
    • https://www.youtube.com/watch?v=iHzzSao6ypE&t=8s
  • Responsibility for Crashes of Autonomous Vehicles: An Ethical Analysis
    • Hevelke A, Nida-Rümelin J (2015) Science & Engineering Ethics, 21:619–630
    • Ludwig-Maximilians-Universität München