Controlling imperfect robot swarms

Robot swarms are incredibly fragile against imperfections.

World-class roboticists agree that one of the fundamental challenges in robotics is on the availability of systematic methods with formal guarantees for the design and control of the swarm's force multiplication, where sensing, actuation, and communication are distributed in space.

However, no matter the approach, control theory, or heuristic, tiny imperfections are amplified throughout large numbers of robots and rapidly erode and make unpredictable the overall performance of the swarm.

Notwithstanding, imperfections can result in surprising complex emergent behaviors such as intricate trajectory patterns of mobile robot swarms.iSwarm questions the current paradigm of fighting imperfections to suppress their “damaging” effects.

Conversely, I propose a rigorous control theory to unleash and ally with imperfections, such as dropouts, delays, and scaling/biasing factors in sensors and actuators, as novel distributed control inputs for taming emergent behaviors.

My paradigm shift requires new ways of analyzing algorithms and their robot integration at the crossroads between algebraic graph theory, network theory, control, and mechatronics for multi-agent systems.To achieve the project's goal, I will: 1) develop a general formulation to characterize the controllability/stabilizability of emergent behaviors with imperfections as inputs; 2) introduce unconventional strategies such as the mismatched Lyapunov functions to engineer emergent behaviors; 3) construct equivalence principles between imperfections and inconsistent shared information to improve the effectiveness of the swarm’s “collective awareness” for fault-recovery algorithms; and 4) demonstrate the control of a state-of-the-art robot swarm in non-lab conditions by exploiting robot imperfections.iSwarm regards imperfections as part of the solution, will inspire innovative research methods and lead new applications of multi-robot systems.