Data and Code for Octopus-Inspired Perceptive Soft Tentacles for Versatile Coordinated Manipulation. Yufeng Wang et al.
Description
This project is associated with the article entitled “Octopus-Inspired Perceptive Soft Tentacles for Versatile Coordinated Manipulation,” published in Device journal by the first author Yufeng Wang. It contains the main raw data and source code used in the study.
Files
Steps to reproduce
The package contains the main raw data, source code, calibration files, model-training files, trained model parameters, and integrated system programs associated with the study. The camera calibration and sensing-calibration files can be used to establish the experimental setup and obtain synchronized vision, pressure, and inductance data. The provided training scripts can be used to train the perception-based shape reconstruction network and the axial mapping model. Based on the trained models and calibrated system parameters, the integrated programs can be used for real-time shape reconstruction, closed-loop control, and trajectory tracking. The current package provides GUI-based manual target adjustment and circular trajectory tracking as representative examples. Other trajectory-tracking tasks and control schemes can be implemented by modifying the target-generation module in the GUI or control program. It should be noted that key parameters, including camera indices, calibration matrices, marker offsets, pressure-channel settings, valve-output channels, PID gains, voltage limits, and trajectory parameters, should be adjusted according to the actual prototype, hardware configuration, and experimental setup. For demonstration tasks involving the robotic arm, the current version uses a teaching-and-playback strategy. The motion trajectories and key position-planning information of the robotic arm were first recorded through manual teaching. During the demonstrations, the program identified different operation stages by analyzing changes in the inductance signals, including both signal amplitude and derivative information. Based on this process recognition, the system triggered the corresponding motion or manipulation actions to complete the demonstrated tasks. Additional experimental details, quantitative results, and supporting data are provided in the main text and supplemental information of the associated paper. For further reproduction of this work, please contact the first author (wyf2021@mail.ustc.edu.cn) or the corresponding author if any questions arise during implementation.