Military Working to Make Its Autonomous Technology Smarter

The Army has autonomous resupply trucks that can be operated by remote control or in convoys in “leader-follower” mode. (Source: https://www.autonews.com/shift/military-working-make-its-autonomous-technology-smarter)
August 13, 2019 | Source: Automotive News, Danielle Szatkowski, 21 July 2019

Keeping soldiers safe continues to be the main reason the military enlists unmanned vehicles into its ranks, especially for resupply missions.

The U.S. Army has been tinkering with ground robotics and automated technologies for decades, yet there are still no real systems in the military field that are fully autonomous. Nevertheless, the Army wants to finish what it started, so it plans to accelerate its use of autonomous defense technologies.

For now, rather than fully autonomous systems, there are semi-automated solutions that require having a soldier on hand, whether it's to monitor the technology or interact with it.

Autonomy Challenges

Engineer Bernard Theisen of the U.S. Army CCDC Ground Vehicle Systems Center in Warren, Michigan, says the U.S. military faces challenges similar to those of commercial developers when it comes to autonomous vehicles.

"The biggest challenge deals with perceiving the world and then processing the data," he said. "For example, most humans take their driver's test when they're 16 or 18 years old, which is like 16 to 18 years of learning how vehicles and driving rules work. Now, we're trying to program a robot to take in all that information and figure it out. What happens is you end up with these edge cases or corner cases that the robot doesn't know how to handle, resulting in system failure."

"This is a significant thing that humans can do," Theisen said. "When we're engaged in a situation we don't understand, we usually build our own solution."

Another challenge that the U.S. military faces deals with testing vehicles and getting them to successfully adapt to changing environments. Theisen said the military will conduct tests in snow and ice, but it must first let the robot know that it's in snow and ice mode.

Robots rely on algorithm planners developed by humans. But if they encounter a situation not covered by the algorithm, they won't know what to do. For example, Theisen said, a self-driving car doesn't know how to react to a person in a wheelchair trying to chase ducks off a road.

"Right now, the intelligence is just not there for robots to smartly select between planners," Theisen said.

Communities: