深度强化学习下的管道气动软体机器人控制

打开文本图片集

中图分类号： TH22 文献标志码：A DOI：10.13338/j.issn.1674-649x.2025.02.008

Pipe pneumatic soft robot control based on deep reinforcement learning

JIANG Yufei1 ， ZHU Qixin2，3，4

（1.School of Electronic and Information Engineering，Suzhou University of Science and Technology，Suzhou 215009，Jiangsu，China;2.School of Mechanical Engineering， Suzhou University of Science and Technology，Suzhou 215009，Jiangsu，China;

3. Jiangsu Province Intelligent Coexisting-Cooperative-Cognitive Robot Engineering Research Center，Suzhou 215009，Jiangsu，China;

4.Suzhou KeyLaboratoryof Coexisting-Cooperative-Cognitive Robot Technology，Suzhou 215O09，Jiangsu，China）

Abstract In complex pipeline environments，soft robots are more suitable for operational tasks compared to rigid robots. However，due to their infinite degrees of freedom and nonlinear deformation characteristics， the control of soft robots posed a significant challenge. To address the dynamic bending motion control of pipe pneumatic soft，a dynamic model was developed based on their deformation characteristics，and a predictive reward-deep deterministic policy gradient （PRDDPG） algorithm was proposed. This algorithm was applied to achieve continuous motion control，enabling the design of an autonomous motion controllr for dynamic bending. The experimental results demonstrate that the PR-DDPG algorithm effectively controls the autonomous continuous motion of pipe pneumatic soft in three-dimensional space，allowing their front ends to reach target positions and orientations. Compared with the deep deterministic policy gradient （DDPG） algorithm，the convergence time of PR-DDPG is reduced by approximately 17% ，and the reward value is improved by about 20% . The PR-DDPG algorithm improves the continuous motion control capabilities of pipe pneumatic soft.

Keywordspipeline soft robot; motion control; deep reinforcement learning; depth deterministic policy gradient algorithm

0引言

由于在有限空间内操作的独特性，管道软体机器人在工业检测领域的应用主要存在2个问题[1-3]：一是管道软体机器人运动性能较低，无法达到机器人检测功能所需要的运动速度；二是由于具有较高自由度和非线性变形的特点，管道软体机器人的控制极具挑战性。（剩余14793字）