Adaptive Impedance Control with Neural Networks and Variable Stiffness for Intelligent Robotic Arm Manipulation

Abstract

Affected by artificial intelligence and automation technology, robotic arms have become increasingly widespread. However, traditional impedance control algorithms have been unable to adapt robotic arms to increasingly complex environmental changes. Therefore, an adaptive impedance control algorithm based on variable stiffness, neural networks, and motion trajectories is developed. Then, it is used for intelligent control of robotic arms. This method employs an impedance relationship model and dynamically optimizes inertia, damping, and stiffness parameters through PD algorithm. The physical quantities such as joint angular acceleration and end displacement are input into the neural network, which outputs key stiffness and interference to estimate the unknown environment in real-time. An adaptive trajectory generation module is designed and combined with contact force feedback to achieve collaborative optimization. The experimental results showed that in terms of contact force with the environment, the error rate between the contact force controlled by this algorithm and the expected value was about 2%. In terms of joint position, the error between the selected joint position in the experiment and the expected value was kept within 0.5%. When running under the motion trajectory generated by the algorithm, the error of each parameter value was less than 1%. The adaptive impedance control algorithm can significantly improve the stability, accuracy, and safety of robotic arm control. The research results have potential applications in improving the environmental adaptability of robotic arms.