Magnetron fish-like microrobots enable efficient learning of complex movements

On May 8th, Xu Sheng and Xu Tiantian’s research team from the Intelligent Bionic Research Center of the Institute of Advanced Technology, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, proposed a set of complex motion learning control methods for miniature fish-like magnetically driven robots.

Through width learning network training, the research team obtained the relationship between the controllable magnetic field change and the various action primitives of the fish-like robot, and realized the complex motion of the fish-imitation robot, and the method does not require complex parameter adjustment, has excellent robust stability, and ensures that the motion process is not affected by external disturbances. The results were published in IEEE Transactions on Cyberneics, an authoritative journal in the field of intelligent control.

In the study, Xu Sheng, associate researcher of the Shenzhen Advanced Institute of the Chinese Academy of Sciences, was the first author, Xu Tiantian, researcher of the Shenzhen Institute of Advanced Institute of the Chinese Academy of Sciences, was the corresponding author, and the Shenzhen Advanced Institute of the Chinese Academy of Sciences was the first unit. 

Due to its reasonable configuration and small scale, miniature fish imitation robots can operate more flexibly in complex and narrow spaces, and have great application potential in small-scale operations such as micropore exploration and targeted therapy. However, due to the strong nonlinear influence between the magnetic field and the robot’s motion, it is very challenging to control the robot’s trajectory according to the required trajectory.

In addition, in complex scenarios, such as in the human body, the accurate coordinates of the ideal target trajectory are often inconvenient to obtain, which limits the application of tracking control strategies. Therefore, it is very necessary to package the underlying motion of the microrobot into basic movements such as straight walking, right angle bending, S-shaped bend, C-shaped bend, etc., and use these basic movements as the primitives of the high-level motion command library, which is convenient to be called on demand in the subsequent macro motion path planning to reduce the solving complexity of real-time control instructions. In this regard, the research team combined with the width learning theory to train and learn the motion primitives of the magnetron fish-like robot to complete a variety of complex movements.

Structure and motion principle of miniature fish-imitation robot, magnetic drive experimental system Photo provided by the scientific research team

Based on Lyapunov stability theory, the research team designed the basic motion controller of micro-robot with width neural network as the main body, and derived the parameter constraints of the controller network to ensure the stability of robot motion, which greatly simplified the controller parameter training and learning process of different motion primitives. In addition, the research team also proposed a controller network parameter training method with magnetic field parameter changes and robot velocity vector changes as the required data, and users only need to change the type of training data to obtain a variety of motion primitives, and the method also considers the training algorithm of stability constraints, which can ensure that the obtained controller must be stable.

Through simulation and experiments, the research team used the proposed learning control method to obtain micro-robot controllers with various motion primitives such as acute angle bending, J-shaped bend, and S-shaped bend, and carried out the obstacle avoidance motion experiment of fish-like robot. During the robot’s movement, the researchers simulated the complex disturbances that may exist in real scenes by artificially shaking the container and violently touching the robot. It is observed that the fish-like robot directly calls C-shaped bend, S-bend and other motion primitives in a complex environment to achieve efficient obstacle avoidance, and the robot can reach the final designated area by using the proposed method, which verifies the strong immunity of the proposed method.

Xu Tiantian, the corresponding author of the paper, said that the results are in line with the idea of high-level motion command planning, greatly simplify the complexity of real-time control instruction solution, and lay a foundation for the optimal motion planning of multi-machine cluster motion or no reference trajectory of microrobots. The research is expected to be applied to complex motion control in drones, unmanned vehicles and industrial robots. (Source: China Science News, Diao Wenhui)

Related paper information:

Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button