Lightweight physics-based models for the control of fluid-mediated self-assembly of robotic modules

Self-assembling robotic systems form a subclass of distributed robotic systems that undertake the fundamental task of structure formation. These systems build desired target structures by putting their constituting robotic modules together in a distributed and stochastic fashion, i.e., through a self-assembly process. The use of self-assembly as the underpinning coordination mechanism provides powerful means for structure formation across a variety of length scales as well as media. In particular, fluidic media have been shown to be very efficient enablers for small-scale self-assembly. In this paper, we consider a distributed robotic system consisting of multiple miniature robotic modules performing self-assembly in 2D, at the water-air interface. The course of the assembly process in the system culminating in a predefined target structure is shaped by the ruleset controllers programmed on the individual robotic modules, allowing only certain formations and ruling out others throughout the process. Designing control strategies relies heavily on accurate models of the system dynamics. Faithfully modeling such systems and their inter-module interactions involves capturing the hydrodynamic forces acting on the modules using typically computationally expensive fluid dynamic modeling tools. Such computational cost restricts the usability of the resulting models, particularly for the purpose of designing optimized controllers. In this paper, we present a new modeling approach and proceed by employing the resulting model for optimizing ruleset controllers. First, we show how the hardware and firmware of the robotic platform can be faithfully modeled in a high-fidelity robotic simulator. Second, we develop a physics plugin to recreate the hydrodynamic forces acting on the modules and propose a trajectory-based method for calibrating the plugin model parameters. Finally, we employ the resulting model and obtain automatically optimized ruleset controllers for given target structures. (C) 2019 Published by Elsevier B.V.