New Perspectives on Network-Based Modeling and Control of Articulated Robots

Multi-agent systems theory has been widely used in robotics to control robot swarms, where a network defines the agents which share information. However research on network-based modeling of articulated systems is more limited, but has shown promise in various applications, from dynamical system simulations, to physical approximations of an oscillatory model. Taking inspiration from these theories, we consider a single articulated robotic (SAR) system as a network, where the nodes correspond to rigid links in the system, and holonomic constraints connecting links are the edges. Using constrained Lagrangian dynamics, the dynamics are derived in both absolute and relative coordinates using familiar graph matrices and similarities are drawn with multi-agent consensus. Based on the dynamics formulation, a decentralized control framework is developed to control the local configuration of the system. We extend the theory to spatial systems and show the main results still hold. To demonstrate the methods, a SAR system is approximated with a network of physically constrained quadrotors, fixed together with lightweight rods. Time-varying trajectory tracking is demonstrated in simulation and on hardware with Crazyflie quadrotors.