Joint Friction Dynamic Estimation for Robotic Finger Using Novel Fixed-Time Adaptive Model Free Algorithm With ZNN-Based Approximator

When performing delicate operations, robot fingers are often faced with the interference of unknown uncertainties and strong nonlinear friction dynamics, and advanced control strategies are needed to compensate for the uncertainties and nonlinear dynamics in the system. Therefore, this paper proposes a new fixed-time adaptive estimation algorithm (FXT-ZNN) based on zeroing neural network to estimate the joint friction force during grasping. The estimated friction force is useful for the compensation controller to effectively improve the control accuracy and system response speed in diverse grasping application. The algorithm does not rely on the friction model and has the characteristics of fast convergence speed and high estimation accuracy. To verify its performance, FXT-ZNN is compared with gradient descent (GDA), least squares (LSA) and fixed-time radial basis neural network (FXT-RBFNN) algorithms. Under standard conditions, disturbances and noise environments, the IAE of our proposed FXT-ZNN is 0.00074, 0.00098 and 0.00075 respectively, which is significantly better than GDA (0.0607, 0.1396, 0.619), LSA (0.0284, 0.0940, 0.0293) and FXT-RBFNN (0.0023, 0.0033, 0.0024). The results confirm the strong robustness, anti-interference and anti-noise of FXT-ZNN, and the method does not rely on the friction model, making it a powerful method for friction compensation in fine robot finger grasping tasks. The friction estimation of robot fingers during grasping based on this control strategy will lay the foundation for grasping applications that require fine manipulation.