top of page

Autonomous Navy Swarms

David Brown

Wouldn't it be great if all warfare was accomplished with (non-sentient) robots? I like to think so. In this JPL work, we developed both an infrastructure and artificial intelligence behaviors for making small swarms (3-6 boats) that could autonomously engage in common naval warfare games.

Sponsored by the Office of Naval Research, this JPL project was to build an infrastructure and behaviors for enabling robust autonomous execution of common maritime scenarios. I worked primarily on the autonomous behaviors themselves; the set of actions that the individual boats took under certain missions. The design challenges of this project were quite exciting, and included:

  • Developing individual agent behaviors that emerged to a swarm behavior without any explicit group-level planning. That is, each agent was free to conduct its own behavior, and agents were supposed to act based on their best possible picture of the world right now.

  • Operating in noisy communications and high false positive detection rate environments (is that a big wave or a boat?)

  • Multi-objective agents; it wasn't enough to just patrol an area, but instead the agents had to keep an eye out for intruders, follow-up on potential intruders with closer inspection by alternative instruments, escort friendlies, etc.

Probably the funnest part of this project was sitting on the boats, compiling code on my laptop that I was immediately pushing out to the whole swarm, as we wzoomed across the water (at gut-lurching speed!), them faithfully executing the behaviors I had programmed for them; if we crashed, it was potentially my fault, and if we succeeded at an elegant patrol route without a hitch, it was my doing too.


A navy promo video on autonomous swarms

bottom of page