This paper proposes a direct adaptive neuro-fuzzy controller to address the adaptive tracking problem for a class of affine nonlinear multi-variable multi-input (MVMI) unknown systems that are linearizable by nonlinear state feedback. The proposed control scheme uses a recurrent neuro-fuzzy model to approximate the system, which combines an underlying fuzzy model with the approximation abilities of high-order neural networks to produce the fuzzy recurrent high-order neural network approximator. A hybrid control scheme that combines an adaptive feedback linearization term and a sliding mode term improves system performance by suppressing the influence of external disturbances and approximation errors. The existence and boundedness of the control signal is always assured by employing the novel method of parameter modified hopping and incorporating it in weight updating laws making the closed-loop system Lyapunov stable. The case of SISO systems is considered separately by following similar analysis with the more general MVMI case. Simulations performed on well-known benchmarks demonstrate the effectiveness of the proposed control scheme.Ã .The aforementioned expression will be negative by appropriately selecting K r so that K r > N , assuming k k 2 =.k k 2 C ı ' B) We also consider the following alternative tracking signal x d D .sin.2t /, cos.t /, 2/ T with initial conditionwhereas for (69),The disturbances added in (70) were selected as random values in the interval OE 1, 1.The main parameters for the control law (17) and the learning laws (33) and (35) are selected as A D d iag. 4.17, 0.96, 6.59/ K D d iag.143.77, 146.75, 106.64/ DIRECT ADAPTIVE NEURO-FUZZY TRAJECTORY TRACKING 683We also consider the following initial conditions for (69) and (72):x 0 D OE 1.5, 1, 4 T and´d 0 D OE 1, 11, 3 T . The disturbances added in 70 were selected as random values in the interval OE 1, 1. The main parameters for the control law (17) and the learning laws (33) and (35) are selected as A D d iag. 3.31, 0.22, 0.04/ K D d iag.66.97, 130.31, 145.65/The parameters of the sigmoidal that have been used are˛1 D 0.68,ˇ1 D 0.35, and 1 D 0.91. Also, the learning rate is selected appropriately as d f l D 0.058 and the outer boundary for the adaptation of weights as " l i D 250 with Ä l,outer f i