DARPA project reveals one person can control dozens of robots

3 months ago (spectrum.ieee.org)

> U.S. Defense Advanced Research Projects Agency (DARPA), experts show that humans can single-handedly and effectively manage a heterogenous swarm of more than 100 autonomous ground and aerial vehicles, while feeling overwhelmed only for brief periods of time

This will surprise nobody who has watched professional Starcraft players.

  • > feeling overwhelmed only for brief periods of time

    There is something deeply, darkly comedic (depressing?) about the qualitative language here. Primarily the way it simultaneously intersects with modern discourse around wellness, anxiety, and mental health in such a banal manner at the same time as the latent/implicit violence of action (given that the obvious subtext is operating semi-autonomous killing machines).

    • It's DARPA, you're really past the moralizing about war stage here, that's just out of context. I don't see UX experts hand-wringing about the effects of advertising when they're designing their products.

      >discourse around wellness, anxiety, and mental health in such a banal manner

      It's not about "feelings" and that might disturb you, but really very many things should be much less about feelings. A whole lot of "wellness, anxiety, and mental health" isn't about feelings but instead being inside or outside the limits of what a person is capable of handling. Facts-based analysis of work and life and people being too far outside their comfort zone could do a lot for many people dealing with mental health issues.

      DARPA does and obviously _needs to_ study these things. One of the most important areas for this are pilots especially during emergencies. It comes from both directions, designing the machine to be manageable and training the human to manage in exceptional circumstances and _knowing the limits_ of both.

      3 replies →

  • Watching professional starcraft players makes you question if they are human. Their control of vast quantities of units and platoons is unreal at moments.

  • But not everyone is a Professional Starcraft player, even with training.

    Besides, I'd prefer a Supreme Commander interface where patrol points can be added/deleted/moved on the fly while factories produce more into that loop including ferry points along the way. Supreme Commander made me feel it was more about strategy than action count.

    • When controlling swarms of drones in a combat situation your micro will matter more than your macro IRL you won't have one operator do both.

  • Good unit AI for RTS allow for amazing results, and there is so much more control/automation that most RTS games could allow for.

  • > The most common reason for a human commander to reach an overload state is when they had to generate multiple new tactics or inspect which vehicles in the launch zone were available for deployment

    This seems misleading- what they said is that when everything is on cruise control the commander does not feel overwhelmed. But if they have to do some high cognitive load task (like reading statuses) or react to a complex situation the commander will feel overwhelmed, which is bad. We want to be able to react quickly and appropriately to all situations, which we can't do when overwhelmed. Being able to handle dozens of bots in a calm situation is meaningless. We need to staff our bot controllers/monitors/commanders at a level that they can handle those top 3% complex wartime scenarios.

    • >> to generate multiple new tactics or inspect which vehicles in the launch zone were available for deployment

      Following up on GP's analogy, I read this as "human overwhelmed by micro" and "human overwhelmed by macro", which... tracks.

      From my own StarCraft experience, the two most taxing parts of the game - the ones where I could easily get confused and lose track of the battle, or even forget what I was doing and why, were:

      1) Micro, i.e. "generating new multiple new tactics" on the fly, manually controlling a bunch of units, whose survival depended entirely on me being able to do it faster than my opponent.

      2) Macro, i.e. "inspecting which vehicles in the launch zone were available for deployment" and deploying them, while queuing production of new ones - while trying to keep track of the front line(s) and spot potential sneak attacks and overall pay attention to the whole map. "Macro management" is easy when it's the only thing you do - but when there's a battle going on, you end up looking at a different part of the map for a second, every second; it basically becomes a form of "micro", except you're micromanaging your attention.

      In both cases, the source of the overwhelm is the pressure of battle - things are changing so fast that few seconds can decide the fate of the battle, possibly of the overall game - but the battles between peers can drag on for minutes, requiring you to sustain that level of focus for extended time, and keep it split between the fighting and the base management; as there too, few seconds of error can put you at a large disadvantage down the line.

      All this to say - I'm not a soldier, so I might be wrong, but I feel that real-life warfare, at least now, isn't this fast-paced. That may change in drone vs. drone scenarios, but with humans on the ground, I imagine taking it slow and methodical will remain the dominant approach.

    • >For example, sensors collected data on their heart-rate variability, posture, and even their speech rate. The data were input into an established algorithm that estimates workload levels and was used to determine when the controller was reaching a workload level that exceeded a normal range, called an “overload state.”

      Based on this, I also think "overwhelmed" might be editorialized language added by the reporter.

  • It makes me wonder if there could be some sort of lower-cost real-life strategy game with cheap(er) homemade drones eventually, kind of like FPV racing now. I'm not a big RTS person but that sounds really fun.

  • Professional Starcraft players prove that this is possible, but my own experience playing Starcraft indicates it's not all that common.

  • Came here to say this - or a more distant comparison - air traffic controllers. Also control is a tricky term - are they directly controlling or tasking a semi-autonomous robot.

> For instance, in a particularly challenging, multiday experiment in an urban setting, human controllers were overloaded with the workload only 3 percent of the time.

That 3 percent is definitely the part where the innocent people are killed

  • given the performance of the Israelis recently, it may be more like the opposite. they would authorize collateral damage of 300 people to get 1 militant, so their "off-target" ratio could be as high as 99.7%

    Israel does path finding for what the U.S. military can get away with.

Starcraft players presumably not surprised.

But seriously, isn't this just a function of how much babysitting the robots require and how good the UI is for controlling them? I don't see why there should be any fundamental limits here.

Will we ever be able to build a war interface for remote controlled drones that is so good it just feels like an RTS game? Or will latency be an issue.

So what can we do to prepare? Like how do you protect against a drone swarm? Can you hack them from afar? Make them unable to communicate? Anything??

I'd better start carrying fish nets with me. If this is how it's going to be then I'd rather go down trying.

In other news, MANHAAAAAAACKS

  • You make another drone swarm and hit drones with drones. That's how it's currently done in Ukraine with standalone small UAVs and multicopters right now, anyway. Shooting them is hard, net-shooting and shotguns work, but it must be automated, since there are a lot of them. EMP is a fairy tale, because you would also fry everything else, and the most common EMP used in wars so far is an atomic bomb. Jamming may work, if frequencies are known, and there is jamming equipment deployed for that frequency, but it's likely you need to jam video, as to not get operator to the point where they can make decision to be executed autonomously, and perhaps intra-swarm communication as well, in case of retranslators,as with a swarm it's possible to make multiple communication channels to the base, since different units can carry different sets of transmitters, so the best bet is drones hunting drones.

  • > So what can we do to prepare?

    EMP is what my ignorance in electronics suggests. But to be fair, people like you and me have no recourse against drone technology.