Overview: Systems comprised of numerous expendable robots have the potential to perform tasks on large spatial and temporal scales quickly, robustly, and with little to no human supervision. The production and deployment of such systems, known as robotic swarms, is approaching feasibility due to recent advances in computing, sensing, actuation, power, communication, control, and 3D printing technologies. In the last few years, the miniaturization of these technologies has led to many novel platforms for multi-robot applications, including a variety of micro aerial vehicles (MAVs) that could conduct coverage tasks very efficiently and at a relatively low cost. However, it remains a challenge to reliably control arbitrary numbers of autonomous robots in unpredictable environments where prior data is unavailable and both global information and communication are limited or unreliable. Given this challenge, the objective of the proposed work is to develop a scalable framework for modeling, analyzing, and controlling the spatiotemporal dynamics of robotic swarms that are to be deployed in such environments. The designed robot control policies should incorporate stochastic robot behaviors such as random encounters with environmental features and should produce target collective behaviors within a specified degree of confidence. We propose a framework that employs a novel application of random vortex methods, used in the numerical analysis of fluid dynamic models, to derive continuum limits of discrete models of robotic swarms and quantify the difference between these two representations, including in the challenging case where robots follow pairwise interaction rules for maintenance of group structure. We will apply techniques of control and optimization to the continuum macroscopic model of the swarm to synthesize robot control policies for a target collective behavior in a top-down fashion. Our control approach will incorporate computational algorithms for compressive sensing in order to reconstruct scalar environmental fields from sparse robot sensor data and to design efficient strategies for robot data collection. We will apply our modeling, analysis, and control framework to a scenario in which a swarm of insect-inspired MAVs is tasked to pollinate a crop field. In this scenario, we will consider cases in which the environment is initially unknown and there are both static and dynamic features of interest, robot failures, and unmodeled disturbances such as wind and obstacles. We will validate our methodology with computer simulations at varying levels of fidelity and through experiments on a multi-robot testbed with ground vehicles.
|Effective start/end date||9/1/14 → 8/31/18|
- National Science Foundation (NSF): $250,000.00