Using SOM for Drone Formation network:

A self-organizing map (SOM) is a machine learning technique that reduces the dimensionality of a high-dimensional dataset. It's a type of artificial neural network (ANN) that uses unsupervised learning to produce a low-dimensional representation of a training sample's input space. This representation is known as a map. SOMs are unsupervised algorithms, such as k-means clustering and principal component analysis (PCA). SOMs have two stages: ordering and convergence. The algorithm has five stages: Initialization, Sampling, Matching, Updating, Continuation.

The map is applied a regression operation to modify the nodes position in order update the nodes, one element from the model (e) at a time. The expression used for the regression is:

With any distance measure, say Euclidean, the winner of an element is the most similar node in the map. The neighborhood is defined as a convolutional-like kernel for the map around the winner. This lets us  update the winner and the neurons closeby and iteratively attain an optimum fit.

The starting point for the drone formation, represented with neurons, can be a grid or a circle. In the latter case, the som will behave like an elastic ring, getting closer to the stimuli while trying to minimize the perimeter

Implementation and test: https://github.com/raja0034/som4drones