|
Abstract
Controlling
vehicles in a fleet via central command has many advantages over peer-to-peer
networks, including improved route planning and a cheaper implementation,
since the individual cars can remain unintelligent. As cars become more
and more intelligent, there will be a greater need for control, tracking,
path planning, and coordination among vehicles, which is the focus of
our project.
Our goal for this project was to create a working system that integrated
vision tracking, path planning, and modern control for rear-wheel drive
vehicles. Our system effectively tracks four separate vehicles and drives
them in various formations behind a leader car controlled by a human being.
To drive in formation, each of the three following vehicles is given a
reference path to its position in the formation. Updated every half of
a second, the reference path guides the cars to their formation and then
maintains the formation. The paths are calculated as a cubic polynomial
of two variables, parametrized by time. The control algorithm is a linearized
controller with four state variables - x and y position, heading angle
and steering angle, and two output variables - steering and throttle.
The system is implemented in hardware with a firewire camera and four
remote controlled vehicles connected to a PC.
Finally, the LabView implementation has been written in modular method
that lends itself to updates and can easily become a test bed for algorithmic
improvements in the future.
|
