| Nowadays,scientific researchers are becoming interested in medical robots which serve as substitutes for traditional surgical techniques in diagnosing and treating human organs.The magnetic field-driven micro-robot features small invasiveness,wirelessness,and strong driving force.Hence,many research groups have begun their study in this regard.They have focused their attention on the application potential of wireless magnetic-driven micro-robots in various human organs,such as targeted drug delivery,micro invasive surgery,active capsule endoscope,etc.The size of micro-robots also determines its specific application areas.For example,robots of micro/nano-levels can be used for cell and tissue operations,while millimeters and centimeters can be used for organ operations.In the past ten years of research,magnetic-driven micro-robots have witnessed rapid development,from rigid bodies to soft bodies,from the two-dimensional to the three-dimensional,from open loop to closed loop,and from single unit to cluster.However,due to the fragility of organisms,it remains a challenge to make the magnetic driven micro-robot move with high precision,high dynamics,and high autonomy in a biological environment with minimal invasiveness and non-destructive movement.This subject is centered on the design and in-depth research of electromagnetic coil driven system,environmental dynamic analysis system and autonomous navigation control system.First,the noise,accuracy and dynamic response capability of the magnetic field generated by the electromagnetic coil system are considered.In terms of hardware,a power amplifier circuit scheme based on a bipolar power amplifier is proposed.By manipulating the output voltage of the circuit,the current flowing through the solenoid coil is controlled.The latter stage of the current sampling circuit applies a Butterworth filter to filter the current and improve the feedback accuracy.In terms of control algorithm,a new magnetic field control algorithm based on a neural network-based adaptive proportional resonant differential controller is designed to compensate for the inductance effects and temperature disturbances on the generation of magnetic fields,and to improve the tracking ability of magnetic fields of different frequencies.Secondly,in view of the complex environment under the microscope field of view,the limited field of view,and the difficulty of identifying and tracking targets and obstacles,semantic segmentation,image stitching,target detection and tracking algorithm based on deep learning technology are proposed,which is similar to SLAM in the field of autonomous driving(Synchronous positioning and mapping).Among them,semantic segmentation is used for pixel-level segmentation of executable and non-driving areas.Image stitching is used for local map to global map construction.Target detection and tracking is used to identify and track micro robots and dynamic obstacles.Third,a microrobot autonomous navigation and control algorithm based on deep reinforcement learning is proposed,which mainly uses neural networks to construct an autonomous navigation and control model from system observation and input controlling to output regulation.Among them,reinforcement learning adopts the training strategy of multi-agents,and uses the competition and cooperation mechanism between multi-agents to guide the microrobots to learn autonomous navigation from the practice in the environment.Finally,the experimental platform is built,and several experiments of autonomous navigation of microrobots are carried out.Experiments such as trajectory tracking,single target tracking,obstacle avoidance tracking and maze exploration are carried out to verify comprehensive capabilities of the magnetic field driven system and environmental dynamic analysis system and autonomous navigation system proposed in this paper.Experimental results show that the microrobot driven by this system has better adaptability and autonomous navigation capabilities in a complex liquid environment. |
PDF Full Text Request