Research On Navigation Of Magnetic Soft Tentacle Based On Machine Learning

The inherent softness and highly non-linear deformation of the soft continuum robot make it very suitable for complex unstructured environments,and it has been widely used in the biomedical field including targeted drug delivery and interventional therapy.Unfortunately,due to low control accuracy,difficulty in miniaturization,and drive hysteresis,traditional soft continuum robots based on pneumatic and tendon drives are difficult to reach the narrow and tortuous vascular system deep in the human body.The emergence of magnetically actuated soft continuum robots has solved the above problems effectively.Among them,magnetically responsive soft continuum robots made of hard magnetic soft materials have attracted extensive attention of related researchers due to their ease of manufacture and shape programmability.In this paper,we proposed an artificial neural network(ANN)model based on machine learning for the control of medical guidewires used in minimally invasive surgery,which can accurately and efficiently control the deformation of the so-called"magnetic soft tentacles" and navigate it to the target position in the simulated vascular path.Here,we call the ferromagnetic soft continuum robot composed of a magnetically responsive tip at the distal end and a commercial guide wire at the proximal end as the magnetic soft tentacle.The ANN model is trained by the data set extracted from the result of parametric simulation.It can accurately predict and output the corresponding control parameter configuration according to the mathematical representation of the input discretized vascular path.This data-driven control model bypasses the difficult kinematics modeling process,and avoids heavy calculation costs.With the predictive performance of the ANN model,the drive parameters of the magnetic soft tentacle under arbitrary deformation can be easily obtained.First,based on the segmented control,we discretize the continuous vascular path into a series of scattered points and take the offset and deflection angle between adjacent points as the characteristic parameters to describe one certain deformation.Then,the two-dimensional feature matrix representing the entire path trajectory is inputted into the learning model.Thus,the model predicts and outputs the accurate control parameters corresponding to all segments of deformation,including the magnetic field pattern and the deformed section length of the magnetic tentacle.The next step is to collect the training data set according to the design requirements of the learning model.To avoid repeated modeling,we perform the parametric simulation analysis of the magnetic tentacles based on the secondary development function of ABAQUS-Python.And take the length of the magnetic tentacles and magnetic field mode as variables to analyze the deformation under different conditions.Here,we adopt the hyper-parameter search process based on Bayesian optimization to optimize the ANN model.After multiple iterations,it shows good generalization performance on the test data set,reaching a root mean square error level of 0.76%.Finally,we design three experiments to further demonstrate the high-precision positioning and efficient navigation capabilities of the learning model.The results show this control scheme holds great potential to apply in the tortuous and small vascular areas deep in the human body,offers a promising possibility to avoid the risk of magnetic radiation and reduce tissue trauma.