Regression-Based Controller Design Using LQR Data for a Two-Wheeled Gyroscopic Combat Robot

Two-wheeled gyroscopic robots are challenging to balance due to their intrinsic instability and sensitivity to external disturbances. This study proposes a hybrid control approach that combines a linear quadratic regulator (LQR) and a regression-based machine learning model to increase stability and adaptability. The regression model is trained with data collected from an LQR-controlled system to predict the optimal control responses under different conditions. The proposed approach utilizes state-space representation and the Jacobian linearization technique to design and simulate the robot’s performance under both controlled and random disturbance profiles. Simulation results indicate that the proposed method ensures steady motion even in dynamic, uncertain environments and increases resistance to external disturbances. Compared to the conventional LQR control method, the regression-enhanced model dynamically adjusts the control parameters, effectively reducing oscillations throughout the robot’s motion and achieving an average energy saving of approximately 80 %. These results demonstrate that robust, adaptive control in autonomous two-wheeled robotic systems can be achieved by fusing data-driven models with classical control techniques.