Inspiration
We were inspired by NeptuneNorth's challenge to implement a system of multiple actors that are resilient to communiction disruption, as well as Marshall Wace's challenge of 'New Worlds'. That got us thinking, how could we use drones effectively in search and rescue operations, even when the drones cannot communicate with each other?
What it does
Given a target and a starting hub, the hive, our drones will leave the hive and search for the target. The mission is complete when the target location has been reported to the hive and all drones have returned. Drones make their own decisions based on their understanding of the world, including where unexplored land is, how close they are to other drones, and where the last location of others drones is. Our webapp let's users see the current understanding of each drone in real-time, as well as the whole map as it gets explored.
Our drones can opperate in any game - it only requires a hook into the webapp! We demonstrate using bees in minecraft.
How we built it
Our webapp uses React, with C# for the backend. This is where the drone are simulated - for real world use, this processing would be moved onto the actual drone. We then have a game hook, which gathers observations from the game world and translates this into a language the webapp can understand. The webapp takes these observations and sends back what each drone should be doing, which the hook will then actuate in the game. We use the Paper framework for our Minecraft plugin, written in Java.
Challenges we ran into
Choosing and implementing an efficient and feasible search algorithm was difficult! We considered using multi-agent RL to learn a policy, or alternatively using LLMs for high-level decision-making, but we decided that a hand-made algorithm using swarm/boid style rules would be best for our task. Figuring out what those rules should be as well as how to make the algorithm both resillient and resilient was an iterative process, but we managed it in the end.
Accomplishments that we're proud of
We're very proud of our webapp - we think the visuals look great! We're also happy with how smoothly it runs.
What we learned
I think the biggest thing I've learnt is that complex behviour can result even from simple rules - you don't always need a learnt complex function to get job done!
What's next for S.W.A.R.M
Expanding into more games and allowing drones to move between them, and eventually the real world too.