Autonomous Vehicles

By: Rotha Aing

What makes a vehicle autonomous?

• “Driverless”

• Different from remote controlled

• 3 D’s– Detection– Delivery– Data-Gathering

3D’s

• Detection – Reasoning– The surroundings and current conditions

• Data-gathering – Search– From the information search knowledgebase

for purposed actions– What to do next?

• Delivery – Learning– View and record results of actions

Current Approaches

• Fully Autonomous– Taxi-like cars

• Autonomous in closed systems– Monorails

• Assistance System– Environment Sensing– Distance Sensors– ABS

Solution Template

• Sensors: Figure out obstacles around the vehicle

• Navigation: How to get to the target location from the present location

• Motion planning: Getting to the location, getting by any obstacles, following any rules

• Control: Getting the vehicle itself to move

Current Issues

• Technical– Sensors

• Understanding the environment

– Navigation• Know its current position and where it wants to go

– Motion Planning• Navigation through traffic

– Actuation• Operate the correct and needed features

Issues

• Social Issues– Trusting the car

• Getting on public roads• Getting people to go in

– Liability Issues– Lost Jobs

What’s been solved?

• Control

• Navigation

• Some issues of Sensory

Control

• Drive-By-Wire

• Sends messages to onboard computers

• Physical ties are unlinked

• In most current cars

Drive By Wire

• When sensor/trigger is pressed, it sends message to the car to perform the tasks

DBW in Autonomous Vehicles

• Replace the human driver

• Activate the sensors/triggers

• SciAutonics– Servomotors for each gear– Large servomotor with belt drive for steering

Navigation

• Already available

• Combination of:– GPS– Roadside

database

Sensory

• Major issue:– Lack of computing power– “More processors”

• Half completed– RADAR– Laser Detection– Cameras

Sensory Information Issues

• Factors of weather– Dust, rain, fog

• Correctly Identifying an obstacle– Shadows vs. ditches– Shallow vs. deep

• Speed of the vehicle and the speed data can be correctly received

Motion Planning

• Most challenging

• Collision Detection

• Affected by:– Quality of Sensory information– Quality of Controls

• Need for algorithm that can determine movements quickly but also the correct ones

“Road Map”

• Decision Tree (Graph)– With points A and G– Fill in free spots (Configuration Space)– Try to link A to G

• Configuration Space Algorithms– Sampling-based

• Faster, less computing power

– Combinatorial• More complete

Configuration Space

DARPA Challenge

• Defense Advanced Research Projects Agency

• 2004 Desert Course

• 2005 Off-road, mountain terrain

• 2007 Urban Challenge– Collision Avoidance– Obey traffic signs

Stanley

• 2005 DARPA Challenge winner

• Volkswagen Touareg modified with onboard computers

Stanley’s Sensory

• 5 LIDAR lasers

• 24 GHz RADAR

• Stereo camera

• Single-lens camera

Path Analysis

• Built in RDDF (database of course)

• Vehicle predominantly followed the RDDF data

Obstacle Detection

• Machine Learning Approach

• Accuracy value of data is based on how human’s perform

• Slows down when a path can not be found quickly

• Grid of either occupied, free, or unknown spots

Issues with mapping scheme

• Errors in determining environment – 12.6% of areas determined as obstacle was

not

Alice out of challenge

Personal Opinions

• Good progress since the first challenge

• Not until the 2007 challenge will we really know if a fully autonomous vehicle is possible in the near future

• Other approaches more likely to be developed into mainstream before fully autonomous vehicles