Enhanced DQN Framework for Selecting Actions and Updating Replay Memory Considering Massive Non-Executable Actions

A Deep-Q-Network (DQN) controls a virtual agent as the level of a player using only screenshots as inputs. Replay memory selects a limited number of experience replays according to an arbitrary batch size and updates them using the associated Q-function. Hence, relatively fewer experience replays of different states are utilized when the number of states is fixed and the state of the randomly selected transitions becomes identical or similar. The DQN may not be applicable in some environments where it is necessary to perform the learning process using more experience replays than is required by the limited batch size. In addition, because it is unknown whether each action can be executed, a problem of an increasing amount of repetitive learning occurs as more non-executable actions are selected. In this study, an enhanced DQN framework is proposed to resolve the batch size problem and reduce the learning time of a DQN in an environment with numerous non-executable actions. In the proposed framework, non-executable actions are filtered to reduce the number of selectable actions to identify the optimal action for the current state. The proposed method was validated in Gomoku, a strategy board game, in which the application of a traditional DQN would be difficult.

Repetition without Repetition or Differential Learning of Multiple Techniques in Volleyball?

A variety of approaches have been proposed for teaching several volleyball techniques to beginners, ranging from general ball familiarization to model-oriented repetition to highly variable learning. This study compared the effects of acquiring three volleyball techniques in parallel with three approaches. Female secondary school students (N = 42; 15.6 ± 0.54 years) participated in a pretest for three different volleyball techniques (underhand pass, overhand pass, and overhead serve) with an emphasis on accuracy. Based on their results, they were parallelized into three practice protocols, a repetitive learning group (RG), a differential learning group (DG), and a control group (CG). After a period of six weeks with 12 intervention sessions, all participants attended a posttest. An additional retention test after two weeks revealed a statistically significant difference between DG, RG, and CG for all single techniques as well as the combined multiple technique. In each technique—the overhand pass, the underhand pass, the overhand service, and the combination of the three techniques—DG performed best (each p < 0.001).

Sliding repetitive control with adaptive bound estimation for motion systems liable to periodic exogenous signals

Periodic exogenous signals often exist in motion systems, especially those involving one or more rotating elements. These periodic exogenous signals deteriorate the performance of motion systems, and these adverse effects cannot be practically eliminated by straightforwardly increasing feedback control gains due to sensor noise, actuator saturation, and unmodeled plant dynamics. This paper describes a sliding repetitive controller for motion systems subject to periodic exogenous signals. Moreover, an adaptive law for bound estimation is devised to ensure the presence of a sliding motion for both repetitive learning and disturbance observation. The tracking motion system of a disk drive is considered in practice, and a traditional repetitive controller is also implemented for performance comparisons with the proposed scheme. Experimental results are reported in this paper, showing the efficacy of the proposed scheme.

Repetitive Learning Sliding Mode Stabilization Control for a Flexible-Base, Flexible-Link and Flexible-Joint Space Robot Capturing a Satellite

During the process of satellite capture by a flexible base–link–joint space robot, the base, joints, and links vibrate easily and also rotate in a disorderly manner owing to the impact torque. To address this problem, a repetitive learning sliding mode stabilization control is proposed to stabilize the system. First, the dynamic models of the fully flexible space robot and the captured satellite are established, respectively, and the impact effect is calculated according to the motion and force transfer relationships. Based on this, a dynamic model of the system after capturing is established. Subsequently, the system is decomposed into slow and fast subsystems using the singular perturbation theory. To ensure that the base attitude and the joints of the slow subsystem reach the desired trajectories, link vibrations are suppressed simultaneously, and a repetitive learning sliding mode controller based on the concept of the virtual force is designed. Moreover, a multilinear optimal controller is proposed for the fast subsystem to suppress the vibration of the base and joints. Two sub-controllers constitute the repetitive learning sliding mode stabilization control for the system. This ensures that the base attitude and joints of the system reach the desired trajectories in a limited time after capturing, obtain better control quality, and suppress the multiple flexible vibrations of the base, links and joints. Finally, the simulation results verify the effectiveness of the designed control strategy.

The Inappropriateness of English for Specific Purposes (ESP) with Learner’s Goals: A Need Analysis on Communication and Islamic Broadcasting Program

The course of English for Specific Purposes (ESP) is supposed to prepare students for the professional sector, yet the course at IAIN Palopo in Indonesi is designed in general English without any professional input or assessment of the learner's needs. This research is motivated by the Communication and Islamic Broadcasting program students’ complaints of unsatisfaction with the course since it does not meet their needs. This research aims to examine the English needs of students studying in the communication and Islamic broadcasting program. The researchers used a mixed-methods strategy that incorporates both quantitative and qualitative research. The participants in this study were 60 undergraduates and 30 graduate students. Data was gathered through questionnaires and interviews. The data was then analyzed using the comprehensive concept of need analysis proposed by Dudley-Evans & St. John (1998). The results revealed that the majority of students learn English to help them advance in their careers. Their top priority in ESP is to improve their speaking skills, followed by listening, reading, and writing. Due to the repetitive learning method, inappropriate textbook, and short duration, according to the interview results, the students were also unsatisfied with the present ESP course.

Adaptive sliding mode repetitive learning control of the upper-limb exoskeleton with unknown dynamics

This article proposes a new adaptive sliding mode repetitive learning control strategy. The proposed controller can obtain satisfactory position tracking performance in the presence of unknown dynamics and external disturbance. The unknown dynamics parameters of the exoskeleton system can be estimated via an adaptive algorithm, which is used to design the sliding mode control law. Besides, the periodic external disturbance of the system can be compensated by repetitive learning to reduce the tracking error. The stability of the proposed method is demonstrated rigorous by the Lyapunov theory. Using an upper-limb exoskeleton model, simulation results demonstrate the effectiveness of the control strategy. The proposed method has a better control performance than other methods.