This project is actively under development. Features and documentation will be updated regularly.
MARL-Dyson simulates resource optimization using multi-agent reinforcement learning. It models autonomous agents optimizing their positions around a central energy source, inspired by the concept of a Dyson swarm.
The energy distribution is generated by creating a uniform random field across a spherical coordinate grid, with values between 0 and 1. A Gaussian smoothing filter is then applied to this random field, creating continuous regions of varying energy levels. The smoothing parameter controls the transition gradient between these regions. The final distribution is normalized to ensure all values remain in the [0,1] range.
The initial development of MARL system is based on the class definition of the swarm agent, where coordinate location, energy collection, directionality, and movement is defined. This is a simple implementation of the swarm agent rule set, and more will be experimented with in the future. These future experiements include a ruleset introduction where only a single agent can occupy a coordinate location at a time, thus limiting movement and generating a more dynamic environment.