A Lightweight Learning Method for Stochastic Configuration Networks Using Non-Inverse Solution

Stochastic configuration networks (SCNs) face time-consuming issues when dealing with complex modeling tasks that usually require a mass of hidden nodes to build an enormous network. An important reason behind this issue is that SCNs always employ the Moore–Penrose generalized inverse method with high complexity to update the output weights in each increment. To tackle this problem, this paper proposes a lightweight SCNs, called L-SCNs. First, to avoid using the Moore–Penrose generalized inverse method, a positive definite equation is proposed to replace the over-determined equation, and the consistency of their solution is proved. Then, to reduce the complexity of calculating the output weight, a low complexity method based on Cholesky decomposition is proposed. The experimental results based on both the benchmark function approximation and real-world problems including regression and classification applications show that L-SCNs are sufficiently lightweight.

Cascade position- torque control strategy based on function approximation technique (FAT) for flexible joint robots

This work deals with control of rigid link robotic manipulators provided with flexible joints. Due to presence of flexible joint dynamics, additional degrees of freedom and underactuation are developed that would complicate the control design. Besides, model uncertainties, unmodeled dynamics and disturbances should be considered in robot modeling and control. Therefore, this paper proposes a cascade position-torque control strategy based on function approximation technique (FAT). The key idea is to design two nested loops: 1) an outer position control loop for tracking reference trajectory, and 2) an inner joint torque control loop to track the desired joint torque resulted from the outer position loop. The torque control loop makes the robot system more adaptable and compliant for sudden disturbances. It increases the perception capability for the target robot mechanisms. Adaptive approximation control (AAC) is used as a strong tool for dealing with time-varying uncertain parameters and disturbances. A sliding mode term is easily integrated with control law structure; however, a constraint on feedback gains are established for compensating modeling (approximation) error. The proposed control architecture can be easily used for high degrees of freedom robotic system due to the decentralized behavior of the AAC. A two-link manipulator is used for simulation experiments.The simulated robot is commanded to move from rest to desired step references considering three cases depending on the selected value of the sliding mode time constant. It is shown that selection of a large time constant parameter related to the position loop leads to slow response. Besides, one of the inherent issues associated with the inner torque control is the presence of derivative of desired joint torque that makes the input control abruptly jumping at the beginning of the dynamic response. To end this, an approximation for derivative term of the desired joint torque is established using a low-pass filter with a time constant selected carefully such that a feasible dynamic response is ensured.The results show the effectiveness of the proposed controller.

Cascade position- torque control strategy based on function approximation technique (FAT) for flexible joint robots

This work deals with control of rigid link robotic manipulators provided with flexible joints. Due to presence of flexible joint dynamics, additional degrees of freedom and underactuation are developed that would complicate the control design. Besides, model uncertainties, unmodeled dynamics and disturbances should be considered in robot modeling and control. Therefore, this paper proposes a cascade position-torque control strategy based on function approximation technique (FAT). The key idea is to design two nested loops: 1) an outer position control loop for tracking reference trajectory, and 2) an inner joint torque control loop to track the desired joint torque resulted from the outer position loop. The torque control loop makes the robot system more adaptable and compliant for sudden disturbances. It increases the perception capability for the target robot mechanisms. Adaptive approximation control (AAC) is used as a strong tool for dealing with time-varying uncertain parameters and disturbances. A sliding mode term is easily integrated with control law structure; however, a constraint on feedback gains are established for compensating modeling (approximation) error. The proposed control architecture can be easily used for high degrees of freedom robotic system due to the decentralized behavior of the AAC. A two-link manipulator is used for simulation experiments.The simulated robot is commanded to move from rest to desired step references considering three cases depending on the selected value of the sliding mode time constant. It is shown that selection of a large time constant parameter related to the position loop leads to slow response. Besides, one of the inherent issues associated with the inner torque control is the presence of derivative of desired joint torque that makes the input control abruptly jumping at the beginning of the dynamic response. To end this, an approximation for derivative term of the desired joint torque is established using a low-pass filter with a time constant selected carefully such that a feasible dynamic response is ensured.The results show the effectiveness of the proposed controller.

Novel adaptive backstepping control for uncertain manipulator robots using state and output feedback

In this paper, a new adaptive control strategy, based on the Modified Function Approximation Technique, is proposed for a manipulator robot with unknown dynamics. This novel strategy benefits from the backstepping control approach and the use of state and output feedback. Unlike the conventional Function Approximation Technique approach, the use of basis functions to approximate the dynamic parameters is completely eliminated in the proposed scheme. Another improvement is eliminating the need to measure velocity by means of integrating a high-order sliding mode observer. Furthermore, utilizing the Lyapunov function theory, it is demonstrated that all controller signals are uniformly ultimately bounded in the closed-loop form. Lastly, simulation and comparative studies are carried out to validate the effectiveness of the proposed control approach.

A Computational Strategy of Variable Step, Variable Order for Solving Stiff Systems of Ordinary Differential Equations

This research study focuses on a computational strategy of variable step, variable order (CSVSVO) for solving stiff systems of ordinary differential equations. The idea of Newton’s interpolation formula combine with divided difference as the basis function approximation will be very useful to design the method. Analysis of the performance strategy of variable step, variable order of the method will be justified. Some examples of stiff systems of ordinary differential equations will be solved to demonstrate the efficiency and accuracy.

Minibatch Recursive Least Squares Q-Learning

The deep Q-network (DQN) is one of the most successful reinforcement learning algorithms, but it has some drawbacks such as slow convergence and instability. In contrast, the traditional reinforcement learning algorithms with linear function approximation usually have faster convergence and better stability, although they easily suffer from the curse of dimensionality. In recent years, many improvements to DQN have been made, but they seldom make use of the advantage of traditional algorithms to improve DQN. In this paper, we propose a novel Q-learning algorithm with linear function approximation, called the minibatch recursive least squares Q-learning (MRLS-Q). Different from the traditional Q-learning algorithm with linear function approximation, the learning mechanism and model structure of MRLS-Q are more similar to those of DQNs with only one input layer and one linear output layer. It uses the experience replay and the minibatch training mode and uses the agent’s states rather than the agent’s state-action pairs as the inputs. As a result, it can be used alone for low-dimensional problems and can be seamlessly integrated into DQN as the last layer for high-dimensional problems as well. In addition, MRLS-Q uses our proposed average RLS optimization technique, so that it can achieve better convergence performance whether it is used alone or integrated with DQN. At the end of this paper, we demonstrate the effectiveness of MRLS-Q on the CartPole problem and four Atari games and investigate the influences of its hyperparameters experimentally.