Robots that surround, ambush and cooperate

The Theory of Mind for Robots, HUMAC

Researchers from Duke and Columbia universities have just taken a very interesting step in robotics. They created Humac, an approach that teaches robots to collaborate with each other like humans and using a cognitive ability typically of ours, theory of mind.

The idea is that robots are able to foresee the movements of other robots, both companions and enemies and the result is robots that surround, ambush and cooperate with an impressive success rate of 84%, Humac or human-guided multi-agent collaboration, allows a single human operator to engage entire teams of robots in collaborative strategies.

Unlike trial and error learning that consumes time and resources, HUMAC uses specific interventions at strategic moments exactly like a coach guiding his team during a match, this approach helps machines understand the collective plan by anticipating both the actions of allies and the reactions of opponents as if they were reading the intentions of others in real time.

A technology highly appreciated by the military.

To test its effectiveness, the scientists placed the mini robots in a dynamic scenario similar to a children's game like catch or hide and seek. For this, they set up a mini arena with some obstacles and limited vision, without the HUMAC. The capture rate was just 36% with just 40 minutes using the HUMAC system the robots learned to cooperate and surround capturing the targets raising the success rate to an impressive 84%.

Even in real tests with ground vehicles the rate remained at 80% the possibilities are enormous, imagine swarms of drones coordinated in real time search and rescue broadcasts after natural disasters, covering areas efficiently without overlapping or even military missions that unfortunately will be the most common, where infantry robots could act as a team, planning ambushes and tactical encirclements without depending on direct human operators.

HUMAC opens the doors to robberies that not only obey, but act in a coordinated autonomous manner and with collective intelligence, and researchers are already planning to scale the system for larger groups in more complex tasks where collective intelligence will be the key.

We are going to reflect a little here, by teaching robots to think like we are giving them social tools that until yesterday were exclusive to humans and some animals, but that opens up a deep dilemma, when the machine learns to cooperate, anticipate and act in a group, is it still under our total control or are we in practice creating collective systems that in the future will be able to act for their own objectives, especially in military environments where ethical lines are already blurred.

Of course, all this is still relatively rudimentary, but if we think about the evolution of technology in the future we will not see robots only at the individual level but rather artificial collective minds and then, where will all this end?