Brain-Like Chips Could Let Drone Swarms Coordinate on Tiny Power

A team of six researchers, led by Linqi Wang, posted a paper on September 12 describing how a group of machines can figure out how to work together — without a human first writing down detailed rules for how each machine behaves. The goal is a 'Nash equilibrium' (a truce where no machine gains anything by changing its own move alone), the same math that describes splitting chores at home, routing traffic, or dividing a shared bill so nobody feels cheated.

The first trick is learning from data instead of models. Normally, engineers must carefully measure a drone or satellite, build an accurate simulation of it, then design instructions. Here, the instructions are computed straight from noisy recordings of what the machine actually did — including unpredictable disturbances like wind or gravitational nudges. That saves months of careful measuring, and it works even when the machine's insides are a mystery.

The second trick is where those instructions run. Instead of standard AI chips that burn power constantly, the team designed the instructions to run on 'spiking' processors — chips that stay silent until a signal crosses a threshold, then fire a brief pulse, much like neurons in a brain. Because they are mostly idle, they can use far less battery. The authors offer a mathematical guarantee: after a short settling period, the machines land close enough to the ideal joint plan — not perfect, but provably within a small error margin.

To show it off, they simulated a group of spacecraft rearranging themselves into a new formation. It worked on paper. The honest catch: this is a dense math paper with computer simulations only. No chip was built, no drone flew, and 'close enough' is baked into the result. Expect years, not months, before anything like this reaches a product — but the direction is clear.