Distributed cooperative transportation of an elastic object using omnidirectional mobile robots

Abstract This paper presents a novel framework for the cooperative transport of a highly elastic membrane using multiple omnidirectional mobile robots, including scenarios involving additional payloads. Since model-based control approaches have enabled very versatile schemes for the transport of rigid bodies in previous works, this is also pursued for the transport of highly deformable objects, where two complementary object models are proposed. First, to capture large deformations and geometrically nonlinear effects arising during transport, a data-driven surrogate model based on a neural network is employed to predict quasi-static pretensioning forces for desired membrane deformations resulting from different payloads. Since the surrogate model is trained on quasi-static equilibrium configurations, it cannot predict the transient membrane dynamics and is therefore unsuitable as a prediction model for feedback control. Instead, a simplified multibody model is integrated into a distributed model predictive control scheme (DMPC) to account for the dynamic aspects of cooperative transportation. The surrogate is trained exclusively on finite element simulation data and provides force setpoints in real-time, which are then explicitly controlled within a cascaded control structure. The DMPC scheme coordinates the robots to both transport the tensioned membrane along a given reference path and exert the desired tensile forces. The proposed approach is validated experimentally using a custom-built robot platform transporting an elastic membrane along various paths and with different payloads, demonstrating precise tracking performance and satisfactory agreement between desired and achieved object deformations.