Design of power curve for soft gripper based on DQN algorithm

• Main Author: Liu, Shubo
• Other Authors: Wei Lei
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2025
• Subjects: Engineering
• Online Access: https://hdl.handle.net/10356/182453
• Institution: Nanyang Technological University

This dissertation presents an in-depth study on developing and training a soft robotic gripper to optimize its gripping performance while minimizing power consumption. With their ability to handle delicate objects, soft robotic grippers hold great promise in industrial automation and human-robot interaction applications. However, attaining the most appropriate balance between gripping performance and energy usage remains essential. The research explores mixing superior sign processing techniques— Exponential transferring average (EMA) and Kalman Filtering—into the education framework to cope with this issue. Those filters are applied to decorate the accuracy of sensor readings and stabilize the managed indicators, thereby improving the precision of gripping maneuvers. The examine systematically evaluates the results of these filtering techniques on the performance of the smooth gripper under three distinct education strategies: (1) a step-smart technique, which incrementally adjusts the control parameters. (2) an abrupt drop strategy, which introduces unexpected adjustments on top-of-things indicators; and (3) a loose exploration approach, allowing the machine to discover numerous configurations autonomously. Experimental outcomes show that the combination of EMA and Kalman filtering improves the accuracy and reliability of gripping predictions and i allows a substantial reduction in energy consumption compared to standard methods. The findings suggest that the proposed filtering-based total framework can effectively mitigate the noise and instability commonly encountered in tender robotic structures, leading to more efficient and sturdy operation. This painting contributes insights into developing power-green robot structures and underscores the significance of adaptive education methodologies in soft robotics. The consequences provide a basis for future studies on optimizing the performance of soft actuators in various complex environments, paving the way for greater sustainability and robot structure.