Abstract: Steady-state drift control can utilize the ultimate dynamic capabilities of unmanned wheeled vehicles, and in certain scenarios, it can also enhance driving safety performance. However, traditional control methods, which are typically built upon precise system models, often yield unsatisfactory results when confronted with complex conditions involving model inaccuracies, multiple constraints, and data loss. To address this problem, a Model-free Adaptive Control Algorithm based on Dynamic Fuzzy Neural Network under constraints (DFNN-cMFAC) is proposed to achieve data-driven control optimization. Firstly, a novel dynamic linearization technique is introduced to equivalently transform the original nonlinear drift model into a dynamically linearized model, thereby eliminating dependence on model precision. Subsequently, incorporating various constraints, a time-varying calculation method for the pseudo-Jacobian matrix (PJM) is designed, leading to the derivation of the model-free adaptive control law. Next, an extended radial basis function (ERBF)-based DFNN is integrated to leverage both online and offline data for training, enabling effective compensation for lost data. Rigorous mathematical proofs are provided for the boundedness of the PJM, as well as the error convergence and stability of the control system. Simulation studies demonstrate the feasibility and effectiveness of the proposed data-driven control method, while real-vehicle tests further validate its practicality.

Key words: unmanned vehicle drift, model-free adaptive control, data packet loss compensation, pseudo Jacobian matrix, dynamic fuzzy neural network