Sgt. 1st Class Ralph Brewer has an odd job that is at the intersection
of past and future warfare—teaching a robot to be a combat
reconnaissance scout.
“An Army field manual can tell you how you should do it,”
he says. “When you get out there and do any type of mission,
it’s not ever the same … There are ways of doing things
the book doesn’t show you.”
Brewer’s work is part of an Army Research Lab’s effort
to instill robots with complex behaviors, thus, making them suitable
for the battlefield. At the lab, a dedicated effort is under way
to create the perfect unmanned reconnaissance scout.
To perform the mission the robots need two things: sensors and
good programming. They must be able to read a situation quickly
and safely, whether it be moving over obstacles or navigating a
course through coming vehicle or pedestrian traffic.
“Those in the military know there is a lot more to being
on a road recon mission than moving from A to B,” says Charles
Shoemaker, head of the lab’s robotics program office. “There
is the traffic on the road. They’re looking for ambush points
and bridges that need to be intact.”
ARL develops robotic platforms, hardware and software, that can
transition into other government research programs, such as the
Army’s Tank-Automotive Research, Development and Engineering
Center. The research lab’s imaging technology, which uses
laser radar to capture local environments, is being used on various
robotic programs at TARDEC. Each of those vehicles is part of Future
Combat Systems.
The laser radar sensors themselves have been honed to increase
their range, accuracy and security. “Earlier generations could
be detected by night vision equipment. We have lowered the emission
frequency so that it can’t be seen,” Shoemaker tells
the audience at a recent industry conference.
Sensors are vital to creating flexible robotic behavior. But the
Army Research Lab is teaching the behaviors. It is starting with
using them to pave the way for convoys and infantry patrols. “There
are many specific tactical behaviors the military wants to integrate,”
Shoemaker says. “We’re focused on one—route reconnaissance.”
The construction of a behavior begins with the textbooks. Instructional
materials are condensed to simple instructions, a series of decisions
a robot can process. Or, as Shoemaker puts it, “that information
can be knowledge-engineered.”
Say a robot is approaching an intersection on a recon mission.
In a structured, methodical way, the machine must know the factors
to which it needs to react. Static features such as walls and roads
are easy. What are more complex are the calculations needed to steer
around pedestrians, observe stop signs, scan suspicious items and
plan an optimal route. Avoiding oncoming traffic requires a system
that can plot the course and speed of oncoming vehicles and determine
if it can proceed without a collision. “We build predictive
dimensions … that help drive the robots’ decision-making,”
Shoemaker says.
However, it takes knowledgeable people to train a robot. “When
we enter an intersection, we know what to look for,” says
Shoemaker. “But anyone who’s taught an adolescent to
drive knows this is not always obvious.” These subtleties
must be taught by professionals, in this case two experienced NCOs
at Aberdeen.
That’s where Brewer, a veteran scout, comes in. “We
go through scenarios with them, changing things,” says Brewer.
Changing conditions, ranging from weather and roadside bomb detection,
can be programmed into the matrix of details on which the robot
needs to train its sensors. Data from unmanned air vehicles can
also be integrated into decision-making.
Often, designs of robots require them to make quick decisions.
The Defense Advanced Research Projects Agency funded the creation
of an unmanned wheeled vehicle, called Spinner, which can take one-foot
obstructions at 20 miles per hour without damage. Any larger obstacle,
however, needs to be avoided.
This requires a perception integration suite that allows the vehicle
to scan and react to such obstacles. Other decisions need to be
made regarding power. When taking a steep incline, the machine can
draw on its batteries for an extra boost.
DARPA also is commissioning robots that navigate on legs, rather
than wheels. That requires a number of different approaches the
machine takes to overcome obstacles.
A gas-engine powered quadruped, called Big Dog, has been constructed
under the agency’s auspices by Boston Dynamics, Inc. The company
builds robots and instills in them the behaviors necessary to navigate
on legs. “The key idea is balance,” says Robert Playter,
vice president of engineering. “They have sensors just as
we have our inner ears.”
When the robot’s leg is raised, internal sensors are converted
into feedback control algorithms that help determine the location
of the machine’s next step. This programming preserves forward
movement and maintains balance. In this way the robot can move autonomously.
“All the things people and animals take for granted we have
to program,” Playter says.
TARDEC has tasked the company to work on a product called Big Dog
Mule, which is programmed to follow a soldier carrying a heavy load.
The ultimate goal is to have the robot follow a soldier up a craggy
mountain path on its own four legs. “The level of performance
you need (from legged robots) doesn’t exist yet,” Playter
says.
Integrating humans and machines is a key priority. The Army is
not interested in something that requires too much specialized training.
“Where we want to go is to make the robot see a platoon member,”
says Jeff Jaczkowski, an electrical engineer for the TARDEC.
Towards that end, DARPA is looking at smaller mechanical animals
for military purposes. A Chihuahua-sized machine, dubbed Little
Dog, is the subject of a solicitation released in late March.
Crucial to the idea behind Little Dog is the ability to navigate
really rough terrain “There’s no way to tele-operate
one of these things,” says Larry Jackel of DARPA’s robotic
initiatives. The robots will have to make those decisions themselves.
The project aims “to create a library of behaviors,”
he says.
Right now, Little Dog can’t surmount a ream of paper laid
flat, Jackel says. Computer models of software designs take minutes
for the simulated four-legged robot to make up its mind and successfully
climb a small, sharp-edged obstacle.
The military researchers are trying to come up with capabilities
before they are asked. They know that when more complex, “thinking”
robots are fielded, troops on the frontline and their commanders
will not want to be limited in urban settings and rough terrain.
Likewise, they are trying to make the sensors efficient, rugged
and small.
“If you have sensors with temperature ranges that we see
in today’s conflicts, you can’t play,” Shoemaker
says. “Everyone is looking for the same real estate on these
vehicles,” he says. “Being smaller is an asset.”
Brewer says the tempo of operations has increased the pace and
interest in developing robotic behavior, especially in the situations
in which robots must be able to function. “We go through scenarios
with them, always changing things,” Brewer says. “If
there’s a new box near the road, an IED or something, we want
them to notice that.”
To the soldiers tasked with teaching the robots, the future means
a safer way for them to do their business. Asked if his job as a
scout would one day be taken over by robots, Brewer was unambiguous:
“It’s coming, no doubt.”