The Anti-Sub Game

Who do you think would come up with better tactics for tracking enemy submarines: A Navy commander or a bunch of online gamers?

ACTUV screenshot

In addition to DARPA’s crowdsourced marine assault vehicle (See "The Wisdom of Cars") another example agency director Regina Dugan’s strategy to overhaul the way the defense department ushers new technologies to the battlefield can be found in an online game.

The agency has been working on a technologically sophisticated unmanned pod called the Anti-Submarine Continuous Trail Unmanned Vessel or ACTUV, which would travel on the surface of the sea to track potentially hostile submarines. It fills a real need. A typical diesel-powered sub might cost $200 million to make but run 10 times that—or $2 billion—to defend against, according to DARPA program manager Rob McHenry. It functions like a Roomba, except instead of vacuuming dust it gloms onto a submarine to constantly track it. Unlike a predator drone, which is piloted from a remote location, the ACTUV relies on software to deploy optimal tactics, and by applying artificial intelligence it learns as it goes. But what would be the best tactics to code into the software? To find out, McHenry put it to the crowd earlier this year in the form of an online game.

He modified an existing game: "Dangerous Waters," and on a DARPA website asked players to come up with ways to track elusive subs that have never been thought of before and confronted with five different configurations to choose from—different combinations of speed and short- and long-range sonar capabilities. Players were asked to eavesdrop on the target: a "politically correct" diesel-electric submarine not specific to any one country; nonetheless it posed a danger since it toted an ambiguous weapon of mass destruction. "It was sort of a terrorist submarine that was threatening the east coast of the U.S." McHenry says, "so the user would have to keep tabs on the submarine so that it could be monitored." When it exceeded a certain threat level, the Navy would destroy it.

The game was posted on DARPA's servers on April 1st and over the course of a month downloaded more than 50,000 times, including one day that saw 12,000 downloads, which crashed the server. It made for a chagrined public affairs officer but McHenry was pleased. When users completed a game, they were asked if they would like to anonymously submit the data file back to DARPA, which led to about 7,000 submissions. Each file was analyzed, and a player earned points for things like the amount of time he was trailed the enemy sub, maximizing propulsion efficiency, sticking to sonar bands that are not associated with environmental risks, while points were deducted for burning too much gas by throttling up and down too much or coming too close to passing commercial ships. "Behaviors and actions that didn’t comply with the rules of the road or posed safety risks to colliding with other traffic were highly penalized," McHenry adds.

After crunching the data, McHenry created five leaderboards then invited a Navy officer expert in tracking subs to play the game. He used time-honored Navy anti-submarine warfare operations tactics, but the highest he placed on a leader board was third. On most he was fifth or sixth. McHenry says that conventional submarine warfare operations maximize the range at which you follow the submarine, so the anti-submarine warfare expert only got as close to the target as he had to. The top public users, however, didn’t put that constraint on themselves. Whether they were operating with long or short-range sensors, they drove as close as they could to the submarine and sat on top; since it was an unmanned craft, there was no fear for loss of life. "That gives you the most tactical control over the maneuvers of that submarine and ability to respond," McHenry says, which the top game players "correctly demonstrated… in hindsight it’s kind of obvious, the closer the better."

One player with the handle "MolonLabe" (In ancient Greek it means "bring it on!") topped the leader boards in four of five categories, but McHenry doesn't know his or her identity. Since game files were forwarded anonymously—McHenry didn't want to scare anyone off from participating—for all he knows the world's best player could be a 10-year-old girl from Israel or Berlin or Saskatchewan. It's very Ender's Game.

The best players broke the rules and in the process may have discovered a better way to track subs from the surface. Says Dugan: "You're looking for the maximum number of folks who can contribute ideas to the process no matter where they come from.

Adam L. Penenberg is a journalism professor at NYU and a contributing writer to Fast Company. Follow him on Twitter: @penenberg.

Related:
Innovator In Chief

Add New Comment

2 Comments

  • Ken Bach

    Maintaining a trail from a "safe distance" was never really a viable option.  They set up the scoring system to encourage the player to achieve contact on the high-frequency sonar sensor (and at least one of any other sensor) at least once every 5 minutes.  The range on the HFS was only about 500yd-1200yd, depending on the ACTUV model.  So unless you wanted to burn a ton of fuel trying to get within that range once every 5 minutes, it made sense just to sit there.  So the idea of staying super-close was more or less dictated by SCS and/or DARPA from the beginning.

    For one of the models, I forget which (maybe the Triton), it was actually counterproductive to get too close to the target, because of the limited vertical field of view of both sonars.  Getting too close would put the sub into your blind spot, so the ideal trail range was actually the max HFS range rather than "as close as you can."

  • Ken Bach

    A 10-year-old girl from Saskatchewan... or a 31 year old attorney from Chicago.  Could be either.  ; )