Two-stage neural network control method for space soft manipulator

Soft robotic arms, with their characteristics of lightweight, low cost, and flexible operation, hold tremendous potential for on-orbit servicing tasks. However, the inverse kinematics modeling and control of soft robotic arms remain challenging. As an alternative solution, the application of data-driven methods to learn numerical models of soft robotic arms has shown some success. Building upon previous research, this paper proposes an end-to-end two-stage neural network control approach and an asynchronous Transformer execution strategy for soft robotic arms. Comparative analysis with single-stage neural networks, traditional backpropagation (BP), long short-term memory (LSTM), and other two-stage methods from prior studies demonstrates that the approach presented in this paper achieves higher control precision. Finally, practical grasping experiments with a physical soft robotic arm validate the feasibility of the proposed method. © 2023 South China University of Technology. All rights reserved.