A multi-layer swarm control model for information propagation and multi-tasking
Abstract
Modeling and control of multi-agent systems is an important problem due to its large variety of potential applications and increasing practical and theoretical challenges. A large
part of inspiration for modeling and control of multi-agent systems originates from the study of natural collective behaviors observed for example in schools of fish, flocks of
birds, colonies of ants and cultures of bacteria. While individuals in these natural swarms are collectively performing complex tasks such as foraging or synchronization, critical information such as predator warnings propagate across the swarm almost instantly and presumably without explicit communication between the individuals. On the other hand, algorithms for multi-agent systems to locate a source or to follow a desired level curve of spatially distributed scalar fields generally require sharing field measurements among the agents for gradient estimation. The dependence on the exchange of data through a communication channel is a hard requirement that might be undesired especially in applications with severe limitations such as underwater robotics. The main contribution of this Dissertation is a Multi-Layer control model composed of an interplay of decentralized algorithms for perception and swarming. In the perception
layer, each agent applies a Principal Component Analysis (PCA) on the relative positions and headings of its neighbors to learn principal properties about the motion and the geometry of the spatial distribution of the surrounding agents. These principal components are then used in the swarming layer where various distributed control laws are designed to balance between achieving a collective task and at the same time allowing critical emerging
signals to propagate to the entire swarm. Within this Multi-Layer model, we contributed distributed control laws for swarms to perform collective source seeking and level curve tracking of scalar fields. These control laws scale to swarms of various sizes and graph structures and do not rely on explicitly estimating the field gradient or explicitly sharing measurements among the agents. Additionally, we contributed a distributed control law that balances between achieving a collective task and at the same time allowing critical signals to propagate to the entire swarm. Through this, we demonstrated implicit information propagation in swarms exhibiting predator-avoidance behavior using only local interactions and without explicit communication or prescribed formations. Moreover, we obtained various stability results reflecting the convergence and robustness of the proposed algorithms. Finally, we validated the proposed model for source seeking, level curve tracking and predator avoidance behaviors through various simulation and experimental results. The proposed control model offers a new method that enables robots with limited resources to perform diverse swarming activities with only local information. Additionally, designing analytical models to understand information propagation will not only reveal natural mysteries but additionally will help to propose multi-tasking control algorithms for robotic swarms that require only very limited or no explicit communication.