Magnetic Microrobot Control Using Neural Fields in Microfluidics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Controlling kinematically complex microrobots in dynamic biomimetic microfluidic environments with magnetic actuation is challenging due to the difficulty in accurately modeling their behavior, especially for microrobots with complex magnetization profiles and soft material composition.
Innovation Solution
A system using a magnetic microdevice with a helical configuration and multi-axis electromagnet controlled by a machine-learning model, specifically an artificial neural network, to generate a non-uniform time-varying magnetic field, enabling autonomous control of microrobots in fluidic environments without explicit system modeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If classical feedback control systems are used to coordinate microrobot behaviors, then control precision can be maintained in simple environments, but the system becomes inapplicable or requires extensive customization for complex uncharacterized environments
Solution Approach 1:
The patent replaces classical mechanical feedback control systems with a machine learning-based control approach. The machine learning model learns optimal control policies directly from environmental interactions, substituting the need for explicit dynamic models and complex feedback controllers, thereby achieving adaptability to complex uncharacterized environments without proportionally increasing control system complexity
Solution Approach 2:
The patent changes the fundamental parameters of the control system by transitioning from model-based control parameters to data-driven learning parameters. The machine learning model adjusts its internal parameters (weights and biases) through training on environmental data, enabling adaptation to complex environments without requiring complex external control mechanisms
2Reliability
If accurate dynamic models of microrobot behavior are created to enable precise control, then control performance improves, but the time and resources required for model development and controller construction increase significantly
Solution Approach 1:
The patent implements self-service control where the machine learning model automatically learns the optimal control policy through direct interaction with the environment. The system serves itself by learning from experience rather than requiring external experts to create accurate dynamic models, thereby maintaining high control performance while dramatically reducing development time
Solution Approach 2:
The patent performs preliminary learning action through a training phase where the machine learning model accumulates experience in the environment before actual control tasks. This preliminary action allows the model to pre-learn optimal control strategies, so that when deployed, it achieves high reliability without requiring time-consuming model development at deployment
3Adaptability or versatility
If complex composite materials and dynamic morphologies are used in microrobots to enhance capabilities, then functional versatility improves, but the difficulty of constructing dynamic and kinematic models increases significantly
Solution Approach 1:
The patent replaces model-based control with machine learning-based control, which naturally handles the complexity of composite materials and dynamic morphologies. The machine learning model learns the effective dynamics directly from observations without requiring explicit mathematical models of the complex materials and morphologies, thereby maintaining functional versatility while avoiding increased modeling complexity
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The system allows for optimized swimming behaviors of microrobots in uncharacterized fluidic environments, reducing the time and resources required for developing high-performance controllers by leveraging deep reinforcement learning to derive actuation policies.
Implementation Method 1
generate, based at least in part on the received sinusoidal current signal, a magnetic field associated with the axis of the electromagnetic coil, and where a superposition of the plurality of generated magnetic fields produces a non-uniform time-varying magnetic field in the fluidic environment
Implementation Method 2
an electromagnet including a plurality of electromagnetic coils each associated with a respective axis, where the plurality of electromagnetic coils each is configured to receive a sinusoidal current signal and generate, based at least in part on the received sinusoidal current signal, a magnetic field associated with the axis of the electromagnetic coil
Implementation Method 3
The magnetic microdevice includes a magnetically susceptible polymer in a helical configuration
Data Source
AI summary
A system for controlling a magnetic microdevice to move within a fluidic environment, comprising an imaging device, a multi-axis electromagnet for generating a magnetic field based on sinusoidal current signals, and a controller for controlling the multi-axis electromagnet based on a machine-learning model, and wherein for training the machine-learning model, the controller is configured to receive images of the fluidic environment and the magnetic microdevice from the imaging device, determine a state comprising position information of the magnetic microdevice based on the images, calculate a measure of performance associated with a first set of sinusoidal current signals previously inputted to the multi-axis electromagnet, wherein the calculated measure of performance is used to adjust parameters of the machine-learning model, and generate based on the position information and the first set of sinusoidal current signals using the machine-learning model, a second set of sinusoidal current signals as inputs for the multi-axis electromagnet.