The mixture of magnetic particles and soft non-Newtonian materials enables outstanding magnetic actuation and malleability (Fig. 2A and Fig. S1). It can serve as a magnetic miniature soft robot to demonstrate related robotic properties individually (Fig. 2B, Figs. S2 and S3, and Movies S1 and S2). The content of magnetic particles plays a profound role in the produced soft robots, including magnetic properties and initial physical properties (Figs. S4 and S5A and B). To work in cooperation with other medical devices or to achieve flexible deployment, the attractive deformability contributes to adapting to various surfaces (Fig. 2C). A weight of 200 g was applied to the magnetic soft robots, and several indentations were left on the dry paper (Fig. 2D). The increment of the magnetic particles led to smaller shape changes and adhesion areas, as shown in the insets of Fig. 2D. Adhesion is important in deploying the robots on various surfaces. To reveal its detailed adhesive forces, we utilized the proposed robots with different material ratios to test their adhesive forces on smooth glass. The increment of the content of magnetic particles contributes to the reduction of the adhesive ability (Fig. 2E). In fact, the increment of magnetic particles means the reduction of the content of mineral oil, leading to a decrease in fluidic properties. It is more difficult to fill the voids or pores of the substrate surfaces and hold surfaces, achieving interlocking and contracting. Thus, the adhesion is reduced (Section S1). Different adhesion forces on different surfaces are also different. We tested the adhesion force on 4 surfaces. The adhesion mechanisms mainly comprise mechanical interlocking and dispersive adhesion in this work. Hence, the physical properties of attached surfaces are essential to the adhesion. For example, rough surfaces tend to form more mechanical interlocking and provide stronger adhesion, which was well proved by the experimental data (Fig. 2F). In those adhesive force experiments, the contact area is constant at 9.6 cm². The ratio of the mass of non-Newtonian materials to the mass of magnetic particles is 1:1. Also, the deployment scenarios can be further extended to multiple materials, where the soft robot can maintain the adhesion for over 1 h (Fig. 2G). For safety considerations, those soft robots can be easily removed with almost no residue (Fig. S5C). These results proved that the robot had good adhesion and deformability, guaranteeing its effectiveness in diverse scenarios. With the help of these properties, the proposed magnetic miniature soft robots demonstrated attractive performances in manipulating objects much larger than the robot. For example, after a soft robot was attached to designated entities, those objects could be actuated by a permanent magnet (height: 30 mm, diameter: 30 mm) (Fig. S6). A 1-g miniature soft robot could actuate or grasp an orange that was 300 times heavier than the robot (Fig. 2H and I). Compared to natural creatures, the proposed robot had an extremely outstanding force output capacity (Fig. 2J). The proposed design is promising in self-robotic executions and provides splendid magnetic empowerment for current medical devices.

The excellent deformability and adhesion pave a new path to deploy the proposed robots by spreading the robots on various surfaces, where the magnetic empowerment and part of the robotic properties can be maintained. Due to their adhesion, the robots can form a thin film (300 μm or less) attached to the glass surface tightly (Fig. 3A and B). Intriguingly, the properties of the formed surfaces are flexibly modified by applying external magnetic fields. With strong magnetic fields (200-mT level), contained magnetic particles are organized, roughening the outboard surface, and reducing adhesion (Fig. 3A and Fig. S7). Without magnetic fields, the fluidic property of the robot smooths the surface and promotes the effect of adhesion (Fig. 3B). Adhesion modification helps to adapt to different assignments. With the help of sufficient force output ability and adhesion, the magnet could quickly turn a page after attaching less than a 1-g robot to the page (Fig. 3C and Movie S3). Several application experiments were conducted to prove the robots’ advantages further. The magnetic soft robot demonstrated excellent adhesion and deformability and was attached well to a large glass ball. Thus, the glass ball could be wirelessly actuated smoothly (Fig. 3D and Movie S4). In addition, when more than 200 g of weight were applied to the robot, the weight could be easily shaken, proving that the robot had sufficient output forces (Fig. 3E and Movie S5). With their soft structure, many classic designs of similar miniature robots can also be achieved for tasks such as cargo delivery. The proposed robot could enclose designed cargo. When a strong magnetic field (over 6 T/m) was applied, the cargo could be squeezed and expelled inside the robot and released (Fig. 3F, Fig. S8, and Movie S6). This suggests an exciting solution to achieve remotely controlled on-demand release tasks. As a more interesting application, a soft robot can be utilized to design on-demand release capsules, where the soft robot seals the capsule. The sealed part could be reopened after applying strong gradient magnetic fields (more than 6 T/m) to achieve on-demand release (Fig. 3G and Movie S7).

The proposed robots can also be deployed at the end of a traditional medical catheter to realize more functionalities (Fig. 4A). After attaching the soft robot at the end of a catheter, the catheter is capable of magnetic actuation for efficient guidance and execution of a task [5,19–21]. Figure 4B illustrates the responses of a magnetized catheter to external actuation magnetic fields supplied by a rotating magnet, which facilitates detailed applications. The catheter and the proposed magnetic robot can work cooperatively to achieve more intriguing functionalities. For example, the catheter can function as a motion tracking, and the robot moves along it to realize high-efficiency and precise drug delivery (Fig. 4C and Movie S8). Once it arrives at the designated destination, the adhesion of a robot contributes to its ability to grasp small objects by avoiding complex manipulations (Fig. 4D and Movie S9). Although connected at the end of the catheter, the deformability of the robots still demonstrates their splendid advantages. Serving as a kind of end-effector, the outstanding morphological adaptivity helps to navigate more crowded corners (Fig. 4E). In complex channels, the magnetic miniature soft robot is capable of guiding medical wire or other equipment in vivo under the actuation of external magnetic fields (Fig. 4F and Movie S10). All these operations are possibly accessible in vivo with the help of ultrasound equipment (Fig. 4G and Movie S11). In this section, the utilized NdFeB permanent magnet in all experiments has a height of 30 mm and a diameter of 30 mm (Fig. S6). Its flexible deployment paves a new path to fully exploit magnetic soft robots that can potentially improve the performance of the current medical tools.

It is challenging to simultaneously achieve soft robot reconfiguration, deformability, and sufficient mechanical strength, which is even more difficult on a small scale. Herein, we utilized the strong magnetic fields (60 mT at least) to control the magnetic particle distribution status to modify the stiffness and related physical properties. In fact, given the mechanical properties of the non-Newtonian fluid [31–34], sudden stimuli can be applied to activate resistance to external interactions, providing a valuable alternative to designing related miniature soft robots (Fig. S9). Three hardening strategies are provided based on these properties, including external force stimulus, dynamic magnetic field stimulus, and static magnetic field stimulus. Introducing magnetic particles should not reduce the performance of a functional material performance too much in miniature magnetic robot designs [7,35–38].

Different ratios of soft non-Newtonian material weight to magnetic particle weight ranging from 1:1 to 1:3 were tested. After adding magnetic particles, the non-Newtonian fluid could still perform well, whereas the balls of magnetic materials could bounce almost as high as the original materials (Fig. 5A and Movie S12). Detailed experimental results are provided in Fig. 5B. For an intriguing potential application, we utilized the magnetic soft robot to enclose several liquid cargos and protect the cargo when it is dropped (Fig. 5C). This is a promising method for cargo loading in biomedical scenarios, especially when it is vital to protect the cargo. All these results are the hardening enabled by the non-Newtonian fluidic properties, where external force sudden stimuli were applied. The force stimuli can also be mimicked by applying external magnetic fields. In this way, the non-Newtonian fluid property was activated to resist external forces and interactions; namely, it hardened. Rotating magnets generated slow and mild stimuli at low frequencies (<4 Hz). Sudden and intense stimuli were generated by rotating magnets at high frequencies (>4 Hz). In this experiment, the rotating magnet provided a magnetic field with a density of less than 30 mT and a gradient less than 3 T/m. In Fig. 5D, the designed miniature soft robot was attached to the end of a catheter and actuated by a rotating magnet (3 Hz). Within less than 1 min, noticeable shape changes appeared when the magnetic robot finally left the catheter (Movie S13). This illustrates the deformability and morphological adaptivity of the proposed robots. Under such circumstances, magnetic miniature soft robots can almost perform as in previous research. In contrast, the most crucial property, hardening, appeared under actuation by a rotating magnet with high frequency (>4 Hz). During the whole actuation period, the magnetic robot demonstrated only negligible changes in shape and stayed stable at the end of the catheter for over 5 min (Fig. 5E and Movie S14). Numerous experiments were performed to reveal further the relationship between external stimuli and deformability, and detailed data are provided in Fig. 5F. It was evident that the increment of the external rotating magnet’s actuation frequency reduced deformability, where the shape change ratio was almost zero. The most intuitive reflections of the programmable hardening were the shape change resistance and load capacity. To further reveal the differences in hardening procedures, we put the same weight on the magnetic robot to compare the indentation under different conditions (Fig. 5G). Static hardening is achieved by a strong static magnetic field (100-mT level), where the indentation is the smallest (Fig. S7) [30]. The indentations were clearer without actuation by an external changing magnetic field (Fig. 5G). After applying a rotating magnet (20 Hz, 60-mT level magnetic fields) to actuate those magnetic robots, the indentations were well reduced, demonstrating the effectiveness of the proposed hardening strategy (Fig. 5H). It was easy to apply the required magnetic fields. For more detailed hardening data, we utilized a rheometer to measure the storage modulus of magnetic robots with different magnetic fields. Strong magnetic fields contribute to powerful hardening effects, and the increment of magnetic particles also somewhat enhances the hardening (Fig. 5I and J). The normal rheometer cannot fully reveal related data since dynamic magnetic field-induced hardening leads to vibration and force interactions between robots and external environments. Considering the changes in static magnetic fields and angular frequencies, the provided data still can prove the effectiveness of the hardening strategy (Fig. S14). The selection of proper hardening strategies is critical. For static magnetic fields, the soft robot is actuated in one direction, contributing to greater adhesion in the actuated direction, which reduces its locomotion ability. In addition, static magnetic fields will cause unavoidable shape changes in open spaces, which might lead to detachment between the robot and medical tools. The robot is actuated with proper frequencies in dynamic magnetic fields, contributing to considerable vibration. The accompanying force interactions between the robot and external environmental surfaces reduce the adhesion, enabling the robot to move smoothly. In addition, it also led to fewer shape changes so that the attached robot could work stably with medical tools. Thus, it would be possible to achieve programmable hardening to realize more powerful functionalities for designed tasks.

Different hardening strategies enable different functionalities with various requirements because different hardening strategies lead to magnetic soft robots’ different accompanied responses. When the proposed robot serves as the adhesive surface, its responses to magnetic fields are intuitive. Without magnetic fields, the formed film is smooth and ultrasoft, where a tiny droplet can leave visible indentations (Fig. 6A). Applying a strong static magnetic field (100-mT level), the formed surfaces are uneven and with small burrs, which is a sign of static hardening (Fig. 6B). Thus, it is accessible to realize flexible modification of targeted surfaces. In addition, the outstanding adhesion of the proposed robots tends to cause undesired adhesion on surfaces, preventing smooth navigation. This problem can also be solved by utilizing hardening strategies. Under these circumstances, static magnetic fields or slow-changing magnetic fields contribute to adhesion promotion (Fig. 6C). By using the dynamic magnetic fields (10 to 20 Hz, fast-changing), the proposed robot can achieve smooth, controlled movement on the glass surface (Fig. 6D). More importantly, as a kind of gripper, the soft robot provides sufficient output forces to remove the thorn inserted into the tissue (Fig. 6E and Movie S15), proving the hardening-induced performance enhancement. Hardening strategies bring lots of stiffness promotion so that several almost impossible assignments to those liquid-based robots can be finished. For potential medical applications, we utilized the sufficient output forces of the robot to clear a model embolism in a constrained channel, where the embolism was made of solid Vaseline (Fig. 6F and Movie S16). The miniature robot succeeded in navigating the solid Vaseline smoothly. Furthermore, the soft robot can demonstrate excellent shape adaptivity and stiffness promotion at the same time. It provides an attractive solution to repair the pipe leak and resist strong water pressure (speed: 30 cm/s or more). Based on those findings, the proposed robots bring novel utilization and deployment strategies for various application scenarios, proving the effectiveness and reliability of such a physical intelligence-inspired design.

The soft non-Newtonian material was produced by 3% mineral oil and 97% carbon silicone gel, demonstrating soft structures and attractive non-Newtonian properties. The adopted magnetic particles were NdFeB microparticles (<5 μm) and 300-mesh Fe-Ni alloy particles (Figs. S10 and S11). Magnetic soft robots were produced by evenly mixing the soft materials and magnetic particles at desired weight ratios (Figs. S12 and S13). The maximum weight ratio of magnetic particles to non-Newtonian materials was 4:1. No additional requirements were needed in the manufacturing procedure. The magnetic soft robots composed of NdFeB are blackish, and those composed of Fe-Ni alloy particles are grayish. They are easy to distinguish in experiments. We also utilized a rheometer to measure the related parameters of the magnetic robots (Fig. S14). A cylindrical NdFeB permanent magnet was used to provide the desired magnetic fields (Fig. S6). To generate rapidly changing magnetic fields, the cylindrical NdFeB permanent magnet was fixed on a motor and rotated at different frequencies (Fig. S15). Adhesion force measurement schematic is provided in Fig. S16. A charge-coupled device camera captured all experimental images.

The introduction of magnetic particles is the basis of the proposed magnetic actuation ability of the robots. In principle, magnetic objects are subject to magnetic forces and torques under external magnetic fields. NdFeB permanent magnets were utilized during the experiments to provide sufficient magnetic field gradients for magnetic force actuation. The magnetic force can be theoretically calculated as follows.where V_m, M, and B represent the volume of the magnetized magnetic soft robots, the magnetization of the magnetic soft robots, and the flux density of the magnetic field, respectively. Through the measured data, it was possible to obtain the theoretical forces. In addition, the adhesion and friction forces between the robot and corresponding substrates were crucial to the wireless control procedure. Solid Vaseline was utilized for lubricating, which was applied on the robot surface when needed. Thus, the uncertainties were severe, and related detailed descriptions were highly sophisticated.

To better illustrate the hardening procedure, the magnetic soft robots were regarded as a kind of shear-thickening fluid. It was essentially a dense mixture of mineral oil (<3%), carbon silicone gel, and magnetic microparticles, as shown in related SEM images (Figs. S11 and S13). When the magnetic soft robot was at rest or static, the gaps between those particles were minimal, and the mineral oil could fill the gaps and thoroughly lubricate the particles. Thus, those microparticles inside the magnetic soft robot could move relatively smoothly, demonstrating ultrasoft properties. When applying rotating magnets to actuate those magnetic particles, the gaps between those particles changed. Some of the particles could not be well lubricated because the particles tended to be knocked together. The movement of particles inside the magnetic soft robot was obstructed by solid–solid friction, demonstrating the enhancement of the mechanical stiffness, namely, hardening [31]. Furthermore, the dynamic magnetic actuation contributes to the vibration of the proposed robots, where their force interactions with external environments also spur their non-Newtonian fluidic properties in hardening. For a more theoretical description, we utilized the velocity field u to represent the fluid. The velocity field was governed by the incompressible Navier–Stokes equation [39].where ρ, u, and p represent the density, velocity vector, and pressure, respectively. F_m, F_c, and F_g denote the magnetic force, capillary force, and gravity-induced body force, respectively. λ = λ(x) is the viscosity ratio extended to the entire domain, equal to one for the continuous phase. ε is the viscosity of the continuous phase, i.e., the environment where we conducted robot experiments. The viscosity of magnetic soft robot η(ϑ) depends on the state variable ϑ within [0, 1], which is a scalar field. ϑ denotes the local state of the magnetic soft robot. When ϑ = 1, those particles inside the magnetic soft robot are jammed, indicating hardening. We utilize the Vogel–Fulcher type divergence to approximate the severe thickening as follows [33,34].where A is a dimensionless constant. When the state is at rest, the steady value ϑ_∗ is determined by the local shear stress γ. The description is given below.where γ₀ is the characteristic stress and ϑ₀ is the limiting value, which represents the state in the stress limit and depends on the detailed properties of the magnetic soft robot, such as the packing fraction of the dispersed granules.

For the static magnetic field-induced hardening, when applying static magnetic fields (>100 mT), extra hardening details exist, as shown in Fig. S7. After applying strong magnetic fields, magnetic particles are all magnetized, which can be regarded as a small magnet. Those small magnets are organized as chain-like structures, which is the bone structure of proposed soft robots, promoting stiffness. When those magnetic particles are magnetized and organized, subject to strong external magnetic fields, particles inside the mineral oils tend to be locked since the oil cannot well lubricate those particles. This also contributes to the hardening procedure.

The results are presented as the means ± SD. Without other specifications, all experiments were repeated at least 3 times independently.