Dynamic integral sliding mode control for nonminimum phase reusable launch vehicles and its robustness enhancing design via deep reinforcement learning

Purpose The reusable launch vehicles (RLVs) may experience limited control authority during certain reentry phases, which leads to nonminimum phase behavior of their attitude control systems. This paper aims to present a novel robustness-enhanced dynamic integral sliding mode control (DISMC) method based on deep reinforcement learning for RLVs with nonminimum phase properties. Design/methodology/approach A novel DISMC method is presented to deal with the unstable internal dynamics of nonminimum phase RLVs. The proposed method reframes the attitude tracking issue as a stabilization task for an expanded system along with a dynamic compensator, and the control scheme is capable of suppressing the system’s unstable zero dynamics while maintaining precise tracking of attitude commands. Furthermore, a robustness-enhancing design is developed through the integration of deep deterministic policy gradient (DDPG) algorithm. The incorporation of DDPG allows for the online adaptive adjustment of the control strategy to the varying flight conditions encountered by the RLVs. Findings Numerical simulations indicate that the robustness-enhanced DISMC scheme is adept at stabilizing the zero dynamics of nonminimum phase RLVs, thereby achieving superior tracking performance and exhibiting enhanced robustness under the presence of uncertainties and wide-ranging flight conditions. Originality/value A novel DISMC scheme is proposed to address the precise output tracking and internal dynamic stabilization of nonminimum phase RLVs, and the incorporation of DDPG ensures robust performance of the control method in the presence of uncertainties. The proposed control scheme is expected to generate a reliable way for dealing with RLV’s attitude control or other systems with nonminimum phase properties.