Extended Deep Lagrangian Network for Robotic Arm Dynamics considering Motor Couplings

Accurate dynamic model is fundamental importance to obtain good performance. In recent years, physics-inspired networks have achieved remarkable success in rigid-body dynamics modelling, particularly in robotics. However, this approach typically requires accurate force and/or torque data, which is often unavailable for practical robots. Besides, neglecting the impact of motor friction torque on system dynamics leads to insufficient model accuracy and reliability. This study introduces an enhanced Deep Lagrangian Network (DeLaN) to robotic dynamics modelling considering motor couplings. The proposed framework uses electrical motor signals (current and/or voltage) and robot motion data to create a hybrid model that combines data-driven methods with physical principles. Compared to the existing schemes, the proposed motor-couplings based robot dynamic model is more accurate. Simulations and experimental studies are conducted to validate the proposed methodology on a 6-DOF Universal Robots 10e robotic arm.