Magnetic quadrupole assemblies with arbitrary shapes and magnetizations (Publication Link)
There are three contributions in this project:
1, We introduce quadrupoles as magnetic assembly modules that fundamentally change magnetic particle-particle interactions. The quadrupole modules break the limitation of dipole symmetry, providing a new paradigm for magnetic assemblies from micrometer colloidal particles to centimeter modular robots.
2, We provide a simple step-by-step method for magnetic assemblies with arbitrary shapes and arbitrary magnetizations. Only a few systems can claim to assemble arbitrary designs in two dimensions, including DNA origami, textured mechanical metamaterials, and robotic swarms.
3, Quadrupole assemblies provide a unique approach to synthesize magnetic metamaterials and soft robots. Compared with 3D printing and laser, magnetic soft robotic assemblies can reconfigure to various morphologies through de-assembly and re-assembly in situ, providing a new strategy to design minimally invasive medical devices.
Magnetic cilia carpet with programmable metachronal waves (Publication Link)
In this project, we developed a highly customizable soft robotic system for cilia research. Unlike existing artificial cilia, we provide a simple method to fabricate hundreds of customized magnetic cilia with programmable metachronal waves. This level of system integration and complexity was only accessible with computer simulations before. To prove the capability of this soft robotic platform, we experimentally confirm two major numerical findings of fluidic transport on cilia carpet (Osterman et. al., PNAS, 2011; Elgeti et. al., PNAS, 2013), for the first time. Furthermore, we show the metachronal waves can propel a soft robot inspired by giant African millipede. We believe this robotic platform provides a powerful tool to spark new discoveries in fundamental cilia research, as well as inspiring new soft robotic designs for biomedical applications.
Magnetically active cardiac patches as an untethered, non-blood contacting ventricular assist device (Publication Link)
Cardiac patches are engineered materials and tissues that can restore heart functions by delivering cardiac cells and other drugs locally on the epicardial surface. Recently, new functions have been introduced into the cardiac patches, making it a promising platform for advanced therapies and disease monitoring. However, existing cardiac patches cannot be actuated mechanically and provide significant support to native heart contraction, thus unsuitable to assist the heart’s pumping function. Here, we present a new type of cardiac patches that can potentially function as a ventricular assistive device. These magnetically active cardiac patches (MACPs) can mechanically compress the heart under external magnetic fields and assist heart pumping. Unlike current ventricular assist devices, the MACPs require no tether to connect to the external power sources, significantly reducing the infections rates and complications. In addition, MACPs provide physiologic pulsatile support and require no direct blood contact, further eliminating serious complications, typical for current ventricular assist devices. We show with in vitro experiments that the ejection fraction is up to 65%, which would show complete restoration of heart ejection. We believe this new concept has great potential for the treatment of heart failure and the development of future cardiac devices that can be remotely actuated.
Artificial microtubules for rapid and collective transport of magnetic microcargoes (Publication Link)
Mobile microrobots are expected to transform bioengineering and therapeutics by precisely navigating in microchannels and performing minimally invasive treatments. The functionalities of these magnetic microrobots have grown substantially in recent years. However, delivery at specific target locations remains challenging. Freely swimming microrobots have relatively low speeds, and it is difficult to swim against the complex fluid flows inside the human body. Therefore, catheters remain the most reliable delivery platform, but they cannot be miniaturized to the micron scale because the required pumping pressure is prohibitively high to transport microcargoes, based on Hagen-Poiseuille’s law.
Here, we develop a new microscale biomedical delivery mechanism that overcomes these challenges, offering rapid transport of microparticles without the need of an enclosing lumen to keep particles concentrated together – just like kinesins walking along microtubules, the microcargoes actively walk outside along a thin filament via micro-patterned magnetic stepping stones.
The strong magnetic field gradient close to the micromagnets enables a firm dynamic anchoring and propulsion along the artificial microtubules with an unprecedented speed. The microrobots can even locomote against strong fluidic flows, which we have quantified with detailed experiments and theoretical modeling. We also discovered a collective motion effect, where many small magnetic particles can self-assemble and subsequently propel along the artificial microtubule together. This process marks a major increase in transport above a critical particle density, where particles essentially push each other forwards. Previous theoretical works showed that this collective effect could enhance cytoskeletal transport in cell biology, but experimental evidence was lacking until now. Finally, we demonstrated that this technology is capable of delivering numerous microparticles accurately at very high concentrations to a specific location inside a branching microfluidic channel network.
Since the Hollywood movie “Fantastic Voyage”, mobile micromachines that can navigate inside the body and cure diseases have been a Holy Grail for scientists and engineers. Despite the advances in imaging, materials, actuation, control, and navigation, it is still very challenging due to the high complexity of in vivo environments and low reliability of these processes. By introducing artificial microtubules, we believe these “micro-highways for microrobots” can provide an alternative solution to the free-swimming microrobots, bringing robust biomedical microtransport much closer to reality.
Self-folding soft-robotic chains with reconfigurable shapes and functionalities (Publication Link)
The idea of self-assembly from a simple chain is widely adopted to synthesize structures from nanometer scales DNA/RNA and protein peptides, which are driven by thermal fluctuation, and up to meter-long robotic snakes, which are driven by distributed motors. However, the implementation of this simple idea at the mesoscale (mm~cm) is still missing, due to the negligible thermal energy (~kT) and prohibitively difficult integration and control of miniaturized motors.
In this work, we presented a generalized method to synthesize functional assemblies at the mesoscale by combining magnetic and elastic energies stored in 3d printed soft-robotic chains (MaSoChains). This method overcomes a fundamental challenge in minimally invasive interventions, which is the size of the tools is limited by the inner diameter of the catheter sheath. We show that it is possible to repeatedly assemble and di-assemble programmable structures and devices at the tip of the catheter, and we further explored the unique features and evaluated their potential in various clinically-relevant situations. We believe this method can be further expanded and customized for a wide range of minimally invasive surgeries to provide less pain, faster recovery, and fewer infections for patients.
Scalable high-throughput microfluidic separation of magnetic microparticles (Publication Link)
Reliable and effective removal of magnetic particles from a suspension is a critical step in many emerging high-throughput applications, including blood purification systems, drug delivery micro- and nanorobots, and wastewater remediation. The required throughput for these applications (~100 mL/min and significantly higher) can be orders of magnitude higher than today’s microfluidic separation systems. Much like the gap between test tube synthesis in the lab and batch reactor production in chemical plants, existing microfluidic separation systems cannot be scaled up in parallel to meet these needs. In this submission, we provide a clear path toward scalable high-throughput microfluidic magnetic separation systems with three major contributions:
1. After analyzing existing microfluidic separation systems, we identify that the magnet arrangement is the bottleneck for scaling up the throughput. Consequently, propose a conceptual evolution from “microfluidic-centered” to “magnet-centered” systems, which reshapes the way we design magnetic particle separation devices.
2. We develop a scalable and effective magnetic particle separation method by combining a micromagnet array and a rotating magnetic field. With experiments and matching simulations, we show that magnetic microparticles self-assemble into large clusters and move collectively across the flow, which is about two orders of magnitude higher than gradient field-based separation methods. The system can be further scaled in parallel at low cost.
3. In the literature, many microfluidic separation systems claim to have “ultrahigh throughput”. However, it is not easy to fairly compare them, and some design features are incompatible with each other. We provide a framework to analyze these separation methods and give a step-by-step guideline to increase the throughput, which can be orders of magnitude higher compared to today’s microfluidic separation devices.
Counterfactual rewards promote collective transport using individually controlled swarm microrobots (Publication Link)
Just like swarms of ants can transport a large and heavy object, collective functions and cooperation among individuals have been at the core in the field of swarm robotics (e.g., Kilobots, wheeled robots, fish robots). In this work, we bring individually controlled swarm robots to the micrometer scale by using laser-driven self-propelled colloidal particles. By taking advantage of a counterfactual rewarding scheme during multi-agent reinforcement learning, we are able to assign individual rewards to each microrobot working together. This enables the effective training of an artificial neural network to perform complex collective tasks, including transport of a large rod to an arbitrary position and orientation. We demonstrate through simulation and experiments that the system is flexible, versatile, robust to strong thermal and environmental noise, robust to malfunctioning units, and can be easily adapted for many tasks and applications. This work presents a major breakthrough at the intersection of swarm robotics, microrobotics, and reinforcement learning.
Milestone achievements in this paper:
• Unprecedented System Complexity. While most previously reported microrobot swarms have only a few global system parameters, our laser-controlled microrobot systems demonstrate up to 200 individually controlled microrobots with 600 controllable degrees of freedom.
• From Collective Pattern to Collective Functions. So far, the control of microrobot swarms has focused on pattern formation. In this work, we take a step forward to investigate the collective transport of a large cargo particle. This collective function involves a non-trivial thermal fluctuation, complex interparticle interactions, and even direct surface collisions. All of these are deliberately avoided in previous work.
• Introducing Counterfactual Reward to Swarm Microrobots. Handcrafting the appropriate reward has always been a major challenge in implementing reinforcement learning algorithms. Here we introduce counterfactual rewards, which assign individual rewards within the group of microrobots, to improve the learning performance and overcome the well-known “lazy agent problem”.
• End-to-end Implementation for Versatile Tasks. Despite the potential of multi-agent reinforcement learning, its successful implementation in swarm robots is rare. This is largely due to the difference between computer-simulated training environments and reality. In this work, we have implemented an end-to-end implementation that allows microrobot swarms to train in experiments and learn from real-world physical interactions.