Swarms of low-cost autonomous robots can potentially be used to collect massive amounts of sensor data and transport heavy payloads across long distances. Novel robots for swarm applications are currently being developed as a result of recent advances in computing, sensing, actuation, power, control, and 3D printing. However, there remains a need for theoretically-grounded methods of programming robotic swarms to achieve target sensing and transport objectives under realistic conditions. In many applications, each robot will have highly limited capabilities, undependable communication, no information on its location, and no map of its surroundings. This research project develops a framework for reliably controlling arbitrarily large populations of robots with such constraints. The project focuses specifically on the problem of regulating robot interactions with different types of features that may be present in their environment. These features can be regions of interest that require certain distributions of robots along their perimeters, or they may be objects that require teams of robots to move them to specified destinations. The developed framework applies techniques from the fields of chemical kinetics, social insect behavior, dynamical systems, feedback control, and optimization. It provides a formal way to program robotic swarms for a range of tasks of wide benefit to society, including environmental monitoring, search-and-rescue operations, disaster recovery, automated construction and manufacturing, and even biomedical imaging and drug delivery at the nanoscale.
This project develops a rigorous methodology for feedback control and optimization of robotic swarm population dynamics to produce target coverage and manipulation behaviors in unknown environments using only local sensing and common broadcast information. The approach incorporates stochastic, deterministic, and hybrid stochastic-deterministic models at different levels of abstraction that describe the robots? roles, task transitions, motion, and manipulation dynamics. It includes a novel encounter-based approach to boundary coverage that does not require characterization of encounter rates or knowledge of environmental parameters. New stochastic strategies for cooperative manipulation are developed in part from models of desert ant group retrieval that are based on experimental data using rigid loads and ant force sensors. The control framework is validated through computer simulations and testbed experiments with small mobile robots, each equipped with a multi-degree-of-freedom gripper that is designed as part of this project. Beyond robotics, the project provides analytical tools for a deeper understanding of surface coverage processes in biology, such as protein adsorption, and the behavioral and biomechanical mechanisms that underlie collective transport strategies in social insects.