Driverless vehicles take to the field
Deere leverages rapid advances in sensors and microcontrollers to let its R-Gator operate without a driver.
The rapid evolution of sensors, processors,
and software is making unmanned ground vehicles (UGVs) viable in
applications as diverse as mining and battlefields.
Military vehicles developed by Lockheed Martin and John Deere and mining vehicles made by Caterpillar
have vastly different usage patterns, but they share many commonalities
in electronic controls and sensors. Those controls combine a complex
array of distributed microcontrollers that manage brakes, acceleration,
and steering based on the input from a number of sensors that constantly
build an image of the terrain along the vehicle’s route.
Among the many technologies needed to
move vehicles autonomously or via remote control, the ability to process
complex algorithms quickly is perhaps the most important. Design
engineers are leveraging the advances of microprocessors and memory
chips to handle the many control functions that occur every second.
“In this area of robotics, we’re really
benefiting from Moore’s Law, which gives us the computing power to do
things we couldn’t do before,” said Stuart Moorehead, Manager of Field
Robotics for John Deere Moline Technology Innovation Center. “We also
have enough memory to store maps and large amounts of data.”
Growth in memory sizes makes it possible
to hold relevant map files, GPS data, and instructions, among other
elements. The rapid capacity increases of flash memory also make it more
cost-effective for engineers to house the complex programs that tell
the vehicle when to slow down, speed up, and turn. This software is
among the most important technology for autonomous transportation.
“Software is the big problem; this is
where we solve the challenges,” Moorehead said. “The algorithms are very
complex and very proprietary.”
These algorithms have to be handled at
very high speeds to ensure that the vehicle makes good decisions as it
travels through previously unseen terrain. At the same time, there is a
strong need for error checking to ensure that a minor fault won’t cause
problems that can quickly cascade into serious problems.
The continuing advances of multicore
processors are helping on both counts. These processors provide high
clock rates while they work in lockstep to ensure accuracy.
“Caterpillar uses several dual-core
controllers for the autonomous system. These multiple controllers can be
used to check that each controller is operating correctly,” said Craig
Koehrsen, Autonomous Technical Steward for Caterpillar.
The modules in autonomous vehicles must
communicate with each other continuously. The central controller is
constantly receiving input from a broad range of sensor modules, and it
must send commands to controllers as diverse as steering actuators and
engine or brake control modules.
These multilayered computing structures
are made of smaller controllers scattered throughout the vehicle. There
are also higher-level computing modules that process data from sensors
like cameras and radar.
“We’ve got a distributed architecture,
with one module for perception, reading in data from sensors,” Moorehead
said. “There’s also a module that keeps track of localization
solutions, which include navigation and sensor input like GPS and
odometry. That works with an inertial measurement unit.”
All these control modules rely on a host
of sensors that provide information about the real world. Radar and
lidar often augment IR and visible light cameras. GPS sensors work in
conjunction with inertial sensors that provide odometry information that
helps when GPS signals are lost.
At one time, these sensors used
proprietary technologies. Some had specially designed sensing modules;
others used extensive packaging to meet the harsh demands of vehicles
moving through dusty or rainy areas with wide temperature ranges. But in
the past few years, high-reliability sensors have been used by
carmakers, making commercial off-the-shelf products readily available.
“Sensors are getting more ruggedized;
the volumes of automotive have driven prices down,” Moorehead said.
“We’re really gaining by utilizing technologies that have become popular
in autos such as radar, ultrasonic, and cameras.”
UGV developers have also been aided by
rapid advances driven by other fields such as unmanned aircraft. As
drones have become one of the U.S. military’s most useful weapons,
engineers have developed a number of different sensor technologies that
give users more depth perception and different perspectives than simple
photographs.
“One of the driving factors is the
improvement of lidar, which we’re using as a primary means of
obstruction detection and avoidance,” said Myron Mills, Program Manager
for Lockheed Martin Missiles and Fire Control. “A number of companies
make devices that are smaller and more ruggedized. These new devices
also perform better.”
No comments:
Post a Comment