Search

Adaptive dynamic programming (ADP) is generally implemented using three neural networks: model network, action network, and critic network. In the conventional works of the value iteration ADP, the model network is initialized randomly and trained by the backpropagation algorithm, whose results are easy to get trapped in a local minimum; both the critic network and action network are trained in each outer-loop, which is time-consuming. To approximate the optimal control policy more accurately and decrease the value iteration ADP training time, we propose a nearer optimal and faster trained value iteration ADP for discrete-time nonlinear systems in this study. First, before training the model network with a backpropagation algorithm, we use a global searching method, i.e., genetic algorithm, to evolve the weights and biases of the neural network for a few generations. Second, in the outer-loop training process, we propose a trigger mechanism to decide whether to train the action network or not, which can save much training time. Examples of both linear and nonlinear systems are induced to verify the superiority of the proposed method compared with the conventional value iteration ADP. The simulation results show that the proposed algorithm can provide a nearer optimal control policy and save more training time than the conventional value iteration ADP.

The seam tracking operation is essential for extracting welding seam characteristics which can instruct the motion of a welding robot along the welding seam path. The chief tasks for seam tracking would be divided into three partitions. First, starting and ending points detection, then, weld edge detection, followed by joint width measurement, and, lastly, welding path position determination with respect to welding robot co-ordinate frame. A novel seam tracking technique with a four-step method is introduced. A laser sensor is used to scan grooves to obtain profile data, and the data are processed by a filtering algorithm to smooth the noise. The second derivative algorithm is proposed to initially position the feature points, and then linear fitting is performed to achieve precise positioning. The groove data are transformed into the robot’s welding path through sensor pose calibration, which could realize real-time seam tracking. Experimental demonstration was carried out to verify the tracking effect of both straight and curved welding seams. Results show that the average deviations in the X direction are about 0.628 mm and 0.736 mm during the initial positioning of feature points. After precise positioning, the average deviations are reduced to 0.387 mm and 0.429 mm. These promising results show that the tracking errors are decreased by up to 38.38% and 41.71%, respectively. Moreover, the average deviations in both X and Z direction of both straight and curved welding seams are no more than 0.5 mm, after precise positioning. Therefore, the proposed seam tracking method with four steps is feasible and effective, and provides a reference for future seam tracking research.

In this paper we address a decentralized neighbor-based formation tracking control of multiple quadrotors with leader&ndash, follower structure. Different from most of the existing work, the formation tracking controller is given in one loop without distinguishing the motion control and attitude control by means of the theory of flatness. In order to achieve an aggressive formation tracking, the high-order states of the neighbors motion are estimated by using a proposed extended finite-time observer for each quadrotor. Then the estimated motion states are used as feedforwards in the formation controller design. Simulation and experimental results show that the proposed formation controller improves the formation performance, i.e., the formation pattern of the quadrotors is better maintained than that using the formation controller without high-order feedforwards, when tracking an aggressive reference formation trajectory.

Robot dynamics and its parameter identification are of great significance to the realization of optimal control and human&ndash, machine interaction. The objective of this research is to address the shortcomings of establishing and identifying the self-developed six-degree-of-freedom (6-DoF) collaborative robot dynamics, which leads to a large error in the predicted torque of the proposed robot. A long short-term memory (LSTM) in an optimized recurrent neural network (RNN) is proposed to compensate the dynamic model of the proposed 6-DoF collaborative robot based on the consideration of gravity, Coriolis force, inertial force, and friction force. The analysis and experimental findings provide promising results. The compensated collaborative robot dynamic model based on LSTM in an optimized RNN displays a good prediction on the actual torque, and the root-mean-square (RMS) error between predicted and actual torques are reduced by 61.8% to 78.9% compared to the traditional dynamic model. Results of the experimental applications demonstrate the validity of the proposed systematic error compensation strategy.

Central Pattern Generator (CPG) has the characteristics of strong adaptability and various output forms, which is suitable for amphibious robotic fish whose environment may change at any time. According to the bionic prototype, the Hopf oscillators on ipsilateral and contralateral side were coupled to form a CPG network topology model, which was used to control the steering gear and realize the waveform propulsion of the fins. In MATLAB, parameters such as amplitude and frequency were set for the model to obtain the motion characteristics under five actual gaits. The output curve is consistent with the theory and the conversion flow field is smooth, which proves the correctness of the CPG model and can be used as the target of subsequent simulation. Finally, through ADAMS and MATLAB co-simulation, the output swing angle curve of the controlled object (steering gear) is obtained, which accords with the abovementioned control goal. This guaranteed the effectiveness of CPG model control. Moreover, the centroid and torque of the steering gear changes in the gait process of crawling and steering are especially analyzed, these changes indicate that the steering gear does not lose too much power during the rigid contact between the fin and the ground and prove the practical feasibility of the theory.

Aiming at the problems of “local minimum” and “unreachable target” existing in the traditional artificial potential field method in path planning, an improved artificial potential field method was proposed after analyzing the fundamental causes of the above problems. The method solved the problem of local minimum by modifying the direction and influence range of the gravitational field, increasing the virtual target and evaluation function, and the problem of unreachable targets is solved by increasing gravity. In view of the change of motion state of robot fish in amphibious environments, the improved artificial potential field method was fused with a dynamic window algorithm, and a dynamic window evaluation function of the optimal path was designed on the basis of establishing the dynamic equations of land and underwater. Then, the simulation experiment was designed under the environment of Matlab2019a. Firstly, the improved and traditional artificial potential field methods were compared. The results showed that the improved artificial potential field method could solve the above two problems well, shorten the operation time and path length, and have high efficiency. Secondly, the influence of different motion modes on path planning is verified, and the result also reflects that the amphibious robot can avoid obstacles flexibly and reach the target point accurately according to its own motion ability. This paper provides a new way of path planning for the amphibious robot.