Adaptive Critic Control of Autonomous Underwater Vehicles Using Neural Networks

Abstract

An adaptive critic neural network-based tracking autopilot design for autonomous underwater vehicles (AUV) will be proposed in this paper. The adaptive critic learning scheme consists of an associative search network (ASN), which is implemented by the three-layer neural network to approximate nonlinear and complex functions of autonomous underwater vehicles, and an adaptive critic network (ACN) generating the reinforcement signal to tune the ASN. A proportional gain controller, an ASN and an adaptive robust element, which can eliminate the approximation errors and disturbances, constitute the control law. The ASN and ACN have the same input and hidden layers, and different hidden-to-output weights. The stability of the closed-loop system can be proved by Lyapunov theory. Traditional adaptive critic controllers learn through trial-and-error interactions, however, the proposed tuning algorithm can significantly shorten the learning time by on-line tuning weights of ASN and ACN. The adaptive critic neural network-based controller is simulated for the tracking control of the AUV in 6 degrees of freedom to demonstrate the effectiveness.

Implementation of remotely operated vehicle for direct inspection of reactor pressure vessel and other water-filled infrastructure

“…In the test, every channel has its own conversion formula between the sample values and the actual voltage value because of their different AFE. The conversion formula of channel 1 to 4 which means four control signals to the four propellers, respectively, is shown in Equation (13), the conversion formula of channel 5 to 7 which means three key voltage signals including 3.3, 5 and 12V voltages on the control panel is shown in Equations (14)- (16), respectively.…”

“…According to the requirements of the ROV used for direct inspection of the RPV and other water-filled infrastructure, the underwater robot must have very good depth control precision, in other words, the ROV must suspend at any depth of the reactor pool. In recent years, many depth control methods have been proposed, such as proportion integration differentiation (PID) control, neural networks control, sliding model control and so on [13][14][15][16]. Different control schemes based on PID and fuzzy techniques are proposed in [17] with their performances compared.…”

Neural networks control of autonomous underwater vehicle

“…As far as neural network controller is characterized by robustness and flexibility to control the uncertain nonlinear problems, neural network controller has been chosen to face the dynamics of AUV. Similar studies have been conducted during recent years indicating the benefits of neural networks controller to deal with the AUVs dynamics [2][3][4][5][6][7][8][9][10][11][12][13].…”

“…The OMLPNN has been chosen to approximate the desiredT CTI). Giving a smooth desired trajectory in terms of the earth fixed position and velocity: (9) The trajectory tracking errors in the position and velocity is defined as, e = TI d -TI and e = iJ d -iJ respectively. A filter for these tracking error signals has been proposed using (20).…”

“…The coefficients are vehicle dependent and would need modification if applied to other vehicles. The functional form of the force equations and the coefficients are given in [9]. The model described includes a model of the cross flow drag effects, and a model of the propulsion system.…”