Coordinated Steering and Driving Actuation for Autonomous Vehicle Drifting Using Physics-Guided SCvx NMPC

Autonomous drifting requires coordinated steering and driving actuation near the tire friction limit, where strong tire nonlinearity and rapidly changing constraints challenge control accuracy and real-time solvability. This article proposes an equilibrium-free successive convexification (SCvx) nonlinear model predictive control framework for drift tracking under constant and varying curvature conditions. The front steering angle and rear-axle longitudinal force are optimized jointly subject to actuator, state, and tire-force constraints. A physics-guided MLP residual tire model is introduced to improve rear-tire-force prediction. Online reference generation determines the heading error, yaw rate, and rear longitudinal force targets from path curvature, lateral error, sideslip variation, and rear slip ratio error, eliminating the need for precomputed drift equilibria. SCvx converts the nonlinear predictive control problem into convex subproblems using virtual control, slack variables, and trust regions. Hardware-in-the-loop experiments confirm stable actuator coordination under both test conditions. Under varying curvature drifting, the proposed method reduces lateral error, velocity error, and yaw rate error by 39.2%, 53.7%, and 24.9%, respectively, compared with the Fiala tire model using the same solver. The results demonstrate improved tracking accuracy and numerical robustness for constrained autonomous drift control.