Multi-body dynamics and deep reinforcement learning-based DVA control for wind load suppression in dual-arm live-line working robot

Dual-arm live-line working robots operating on transmission lines are subject to coupled disturbances arising from stochastic wind loads and configuration-dependent inertia variations, which can induce significant roll oscillations and compromise system stability. To address this problem, this article develops a coupled multibody roll dynamics model that explicitly incorporates time-varying inertia and its rate of change under wind excitation modeled by the Davenport spectrum. Based on this dynamic framework, a dynamic vibration absorber (DVA) is introduced to suppress low-frequency oscillations. It is shown that conventional fixed-parameter DVAs suffer from frequency detuning when system inertia varies with manipulator motion. To overcome this limitation, an adaptive parameter regulation scheme is proposed to enable the real-time adjustment of DVA stiffness and damping according to system states. A deep reinforcement learning method is employed to approximate the nonlinear mapping between system dynamics and optimal absorber parameters. Furthermore, the zero moment point is adopted as a stability indicator to evaluate the dynamic balance of the coupled system under wind disturbances. Simulation results demonstrate that the proposed approach effectively suppresses roll oscillations, mitigates detuning effects, and improves robustness under varying wind conditions and time-varying inertia, outperforming conventional fixed-parameter DVA and PID-based methods.