Microrobotic Smartlets

• Microrobotic smartlets are microscale robotic units that integrate minimal structures with rich sensing, actuation, and computation across diverse material platforms.
• They employ varied actuation mechanisms—from magnetic swarms to bubble propulsion—to achieve targeted locomotion, docking, and intervention in constrained environments.
• Emerging designs leverage collective intelligence and embodied physics to enable self-assembly, energy harvesting, and on‐board computing for biomedical and assembly applications.

In the cited literature, microrobotic smartlets designate a heterogeneous class of microscale robotic units and ensembles that combine minimal physical structure with disproportionately rich functional behavior. The term is used for paramagnetic nanoparticle swarms that self-assemble, become conspicuous under clinical ultrasound, and translate near surfaces under rotating magnetic fields (Wang et al., 2018); for physically intelligent micro-robots whose shape, anchoring, or surrounding medium stores control-relevant information (Yao et al., 2022); for modular micro-origami cubes with onboard energy harvesting, CMOS control, optical communication, and bubble actuation (Lee et al., 2024); and for sub-millimeter electronic microrobots that sense, think, act, compute, and communicate with onboard systems for memory, sensing, locomotion, and programmable computation (Lassiter et al., 29 Mar 2025). Across these embodiments, the recurring idea is that useful microscale agency can arise either from explicit integration of sensing, actuation, and digital logic, or from collective and embodied physics that converts local interactions into navigation, transport, assembly, or intervention.

1. Conceptual scope and definitional boundaries

A common source of ambiguity is that “smartlet” does not name one morphology or one actuation modality. In the cited work, it spans field-assembled colloidal swarms, modular electronic microcubes, chemically or optically driven microswimmers, deformable active-matter composites, and larger-scale antecedents of swarm embodiment. In the magnetic ultrasound-guided lineage, a smartlet is “a microscale population that self-assembles on demand, maintains collective behavior under actuation, and can be regathered if dispersed” (Wang et al., 2018). In the modular electronic lineage, the stronger claim is that a “true” microrobot must harvest or carry its own source of energy and its own programmable microcontroller of actuators for locomotion using information acquired from its own sensors (Bandari et al., 24 Aug 2025).

The conceptual range also includes systems whose intelligence is not reducible to explicit onboard sensing. In nematic liquid crystals, physical information is embedded in the director field $n(\mathbf{x})$, in topological defects, and in the multi-stable elastic energy landscape surrounding a rotating ferromagnetic micro-robot (Yao et al., 2022). In reinforcement-learning experiments on self-thermophoretic colloids, hidden flow information is inferred from embodied action outcomes even though the agent state excludes flow measurements (Paul et al., 25 Aug 2025). This suggests that, within the smartlet literature, “intelligence” may reside in digital logic, in morphology, in defect-mediated interactions, or in population-level organization.

A second misconception is that smartlets are necessarily single-body machines. The literature explicitly includes ensembles whose members are individually simple or even non-translating. “Phototactic supersmarticles” are collectives of smarticles confined by an unanchored rigid ring; a single smarticle cannot rotate or translate in the plane by itself, yet the ensemble locomotes by contact-mediated collisions and light-modulated activity asymmetry (Cannon et al., 2017). The Jasmine open-hardware platform extends the same logic to small networked units whose local rules are deliberately coupled to hardware constraints through “swarm embodiment” (Kernbach, 2011).

2. Embodiments and material architectures

The material realization of smartlets ranges from colloidal assemblies to chip-integrated micro-origami modules. The following representative embodiments illustrate the breadth of the design space.

Embodiment	Key dimensions and materials	Integrated function
Magnetic colloidal swarm	Magnetite $(\mathrm{Fe}_3\mathrm{O}_4)$ nanoparticles, average diameter approximately 500 nm	Self-assembly, ultrasound localization, magnetic steering (Wang et al., 2018)
Nematic ferromagnetic micro-robot	SU-8 four-armed structure, thickness $H \approx 25\,\mu\mathrm{m}$, sputtered Ni	Defect-mediated docking, transport, release, assembly (Yao et al., 2022)
Janus photocatalytic microswimmer	Silica cores $0.55$–$4.16\,\mu\mathrm{m}$ with approximately $50/50$ nanoparticle cap	Light-driven propulsion and modular catalyst substitution (Bailey et al., 2021)
COF microswimmer	TAPB-PDA-COF spheres $452 \pm 74$ nm; TpAzo-COF particles $6.97 \pm 17.62\,\mu\mathrm{m}$	Visible/red-light propulsion, loading, OCT/PA theranostics (Sridhar et al., 2023)
Micro-origami cube smartlet	Free-standing micromodules $\leq 1\,\mathrm{mm}^3$ with chiplets and rolled uOSCs	Ambient-power harvesting, communication, buoyancy control, collective docking (Lee et al., 2024)
CMOS electronic microrobot	$210 \times 340 \times 50\,\mu\mathrm{m}^3$ or $(\mathrm{Fe}_3\mathrm{O}_4)$0	Onboard sensing, memory, locomotion, and computation (Lassiter et al., 29 Mar 2025)

These embodiments solve different scaling bottlenecks. Colloidal smartlets exploit large numbers, induced dipoles, and reconfigurability rather than discrete onboard subsystems (Wang et al., 2018). Micro-origami cubes and modular electronic smartlets use folding to increase functional surface area, placing energy harvesters on edges while reserving faces for docking, sensing, or actuation (Lee et al., 2024, Bandari et al., 24 Aug 2025). Fully electronic microrobots use foundry CMOS and post-CMOS lithography to integrate photovoltaics, an optical receiver, a temperature sensor, memory, actuator drivers, and electrokinetic electrodes within a body comparable in size to a single-celled paramecium (Lassiter et al., 29 Mar 2025).

Material choice usually encodes task specificity. The multifunctional polymer route for Janus microrobots binds silica to transition-metal-oxide nanoparticles through silane and nitrocatechol groups, enabling large batches of photocatalytic particles with tunable caps and light response (Bailey et al., 2021). Covalent organic frameworks add large surface areas, structural pores of about $(\mathrm{Fe}_3\mathrm{O}_4)$1–$(\mathrm{Fe}_3\mathrm{O}_4)$2 nm or $(\mathrm{Fe}_3\mathrm{O}_4)$3–$(\mathrm{Fe}_3\mathrm{O}_4)$4 nm, and loading capacity for doxorubicin, insulin, and indocyanine green in intraocular media (Sridhar et al., 2023). Thermo-responsive gelatin capsules embedding zinc-doped iron oxide nanocubes and tantalum nanoparticles prioritize radiopacity, magnetic responsiveness, and dissolvable therapeutic payloads at millimeter scale (Landers et al., 20 Jan 2025).

3. Locomotion and actuation physics

Smartlet locomotion is governed by whichever field, interface, or medium can be exploited most efficiently at small scale. In the colloidal magnetic case, each nanosphere of radius $(\mathrm{Fe}_3\mathrm{O}_4)$5 acquires an induced dipole moment

$(\mathrm{Fe}_3\mathrm{O}_4)$6

and chains rotating synchronously in a field satisfy the torque-balance relation

$(\mathrm{Fe}_3\mathrm{O}_4)$7

At $(\mathrm{Fe}_3\mathrm{O}_4)$8 and $(\mathrm{Fe}_3\mathrm{O}_4)$9–$H \approx 25\,\mu\mathrm{m}$0, these chains aggregate near a boundary into a dense swarm after about $H \approx 25\,\mu\mathrm{m}$1–$H \approx 25\,\mu\mathrm{m}$2 s, reaching area densities of roughly $H \approx 25\,\mu\mathrm{m}$3–$H \approx 25\,\mu\mathrm{m}$4 and translating near surfaces when a small pitch angle is added to the rotating field (Wang et al., 2018).

Cohesive magnetic smartlets use a different magnetic operating point. Self-assembled chain microrobots under a global precessing field balance long-range dipolar attraction against short-range multipolar repulsion. In reduced form, their pair interaction is written

$H \approx 25\,\mu\mathrm{m}$5

with $H \approx 25\,\mu\mathrm{m}$6 in the range $H \approx 25\,\mu\mathrm{m}$7–$H \approx 25\,\mu\mathrm{m}$8. This produces self-bounded clusters that translate above a wall by near-wall hydrodynamics. Cohesion for chains with about three beads was observed for $H \approx 25\,\mu\mathrm{m}$9 at $0.55$0, whereas $0.55$1 led to divergence and $0.55$2 to collapse (Yigit et al., 2019).

Other smartlets are propelled by phoretic, electrochemical, or buoyancy mechanisms. Janus photocatalytic microswimmers use asymmetric reaction fields with slip velocity

$0.55$3

and translational velocity

$0.55$4

Under UV illumination and $0.55$5 $0.55$6, median speeds around $0.55$7 were reported for $0.55$8 silica-based swimmers, with instantaneous velocities spanning $0.55$9–$4.16\,\mu\mathrm{m}$0 (Bailey et al., 2021). COF microswimmers extend this optical actuation into visible and red wavelengths, with TAPB-PDA-COF reaching $4.16\,\mu\mathrm{m}$1 at $4.16\,\mu\mathrm{m}$2 nm and TpAzo-COF sustaining propulsion at $4.16\,\mu\mathrm{m}$3 nm in biological media (Sridhar et al., 2023).

Electrolytic and bubble-mediated smartlets convert electrical power into local gas generation. For the modular cube divers and surface-gliding smartlets, gas production follows Faraday’s law,

$4.16\,\mu\mathrm{m}$4

while bubble pressure is estimated by

$4.16\,\mu\mathrm{m}$5

In the surface-locomoting cube lineage, bubbles of about $4.16\,\mu\mathrm{m}$6–$4.16\,\mu\mathrm{m}$7 generate sufficient pressure asymmetry to tilt a face and produce steps of about $4.16\,\mu\mathrm{m}$8 at roughly $4.16\,\mu\mathrm{m}$9, yielding measured speeds near $50/50$0 on wet glass (Bandari et al., 24 Aug 2025). In the MRI-powered capsule lineage, a submillimeter release hole traps an air bubble as a passive stopper until HIFU removes it, after which acoustic streaming regulates release rate and multi-site dosing (Tiryaki et al., 2023).

Finally, some smartlets do not fight the surrounding flow but exploit it. Ultra-flexible endovascular uprobes use physiological hydrokinetic energy for transport, while uniform magnetic fields deform a soft-magnetic head at bifurcations to bias branch selection. Their body cross-sectional area can be as small as $50/50$1, and advancement velocities in ex vivo rabbit ear vasculature reached about $50/50$2 (Pancaldi et al., 2020).

4. Sensing, imaging, computation, and communication

Sensing in smartlets spans clinical imaging, local scalar measurements, and fully digital onboard instrumentation. In magnetic colloidal smartlets, the most distinctive signal is ultrasound contrast rather than direct optical visibility. The rotating swarm is imaged in $50/50$3D B-mode at $50/50$4 frames per second, and the internal chains modulate acoustic backscatter periodically as they sweep through the yaw angle $50/50$5. Mean pixel intensity rises from approximately $50/50$6 AU in the initial low-density region to approximately $50/50$7 AU in the dense swarm region at $50/50$8 and $50/50$9, with phase-locking to the external field providing a robust localization signature (Wang et al., 2018).

Biomedical smartlets also leverage modality-specific visibility. COF microswimmers can be tracked in real time by OCT in intraocular fluids without added contrast, while indocyanine-green loading enables photoacoustic imaging and hyperthermia. In aqueous humor and vitreous under OCT, TAPB-PDA-COF moved at $452 \pm 74$0 and $452 \pm 74$1, respectively, and TAPB-PDA-COF achieved photoacoustic mean pixel intensity up to about $452 \pm 74$2 at $452 \pm 74$3 nm (Sridhar et al., 2023). Clinically ready magnetic capsules instead use fluoroscopic visibility from tantalum nanoparticles and are tracked at $452 \pm 74$4 fps with DSA roadmaps at $452 \pm 74$5 fps during catheter-based navigation (Landers et al., 20 Jan 2025).

At the opposite end of the integration spectrum, electronic smartlets implement explicit onboard sensing and computation. The CMOS microrobot “that sense[s], think[s], act[s], and compute[s]” integrates photovoltaics, an optical receiver, a temperature sensor, a custom 11-bit CISC processor, instruction memory of $452 \pm 74$6 bits, data memory of $452 \pm 74$7 bits, four 8-bit registers, and four electrokinetic actuators within a body of volume about $452 \pm 74$8 (Lassiter et al., 29 Mar 2025). Its temperature sensing resolution is about $452 \pm 74$9. The modular electronic smartlet lineage uses a custom $6.97 \pm 17.62\,\mu\mathrm{m}$0 nm CMOS lablet of size $6.97 \pm 17.62\,\mu\mathrm{m}$1, a $6.97 \pm 17.62\,\mu\mathrm{m}$2-bit program, differential sensory inputs, and optical programming through Manchester-encoded commands to control face-selective bubble actuation (Bandari et al., 24 Aug 2025).

Communication occupies a similarly broad design space. Micro-origami cube smartlets communicate optically using pulsed micro-LEDs and micro-organic photodetectors, with measured bandwidth of $6.97 \pm 17.62\,\mu\mathrm{m}$3–$6.97 \pm 17.62\,\mu\mathrm{m}$4 Hz and underwater range below $6.97 \pm 17.62\,\mu\mathrm{m}$5 mm (Lee et al., 2024). By contrast, some smartlets substitute embodiment for explicit sensing channels. In self-thermophoretic particles trained by PPO, the state comprises only position and step-distance change, yet the learned policy counteracts hidden flows up to four times the propulsion speed by exploiting the fact that observed displacements encode the joint effect of actuation, advection, and noise (Paul et al., 25 Aug 2025). This suggests that smartlet information processing is not restricted to named sensors and can be distributed across body, substrate, and fluid.

5. Collective intelligence, embodiment, and control

Collective smartlet behavior often arises from minimal local rules. In phototactic supersmarticles, each smarticle is either active or inactive depending on whether a photoresistor exceeds threshold. Because the illuminated smarticle shadows its neighbors, the ensemble acquires an activity asymmetry that biases otherwise Brownian-like motion. The measured mean-squared-displacement exponent is about $6.97 \pm 17.62\,\mu\mathrm{m}$6 for the fully active control and $6.97 \pm 17.62\,\mu\mathrm{m}$7 for the light-directed case, and $6.97 \pm 17.62\,\mu\mathrm{m}$8 of trials drift toward the light source (Cannon et al., 2017). The key point is that no global localization or shared orientation is required.

Embodiment can also provide the control substrate itself. In nematic smartlets, the free-energy landscape around a stationary four-armed ferromagnetic robot contains five recurrent minima and docking modes—dipole-chaining, zig-zag, dipole-on-hill, dipole-in-well, and hybrid—whose existence depends on hybrid anchoring, sharp edges, and defect pinning. Rotation dynamically rewrites this landscape through defect elongation and hopping, enabling cargo docking, transport, release, and even “juggling” with simple magnetic field schedules rather than algorithmic micromanagement (Yao et al., 2022).

Field-driven collectives admit still another control logic: tuning the interaction law itself. Cohesive magnetic-chain smartlets grow up to $6.97 \pm 17.62\,\mu\mathrm{m}$9 chains and show a transition from solid-like ordering to liquid-like internal rearrangements as cluster size increases. Small clusters exhibit bounded fluctuations, whereas for about $\leq 1\,\mathrm{mm}^3$0–$\leq 1\,\mathrm{mm}^3$1 the mean-squared displacement grows at long times after subtracting cluster translation and rotation, reflecting the increasing importance of long-ranged near-wall hydrodynamic advection relative to local magnetic cohesion (Yigit et al., 2019).

Reinforcement learning introduces a formal control layer over similarly constrained physics. Hierarchical PPO has been used to learn topology-specific gaits for multi-link microrobots and then sequence those gaits for chemotaxis under partial observability, using only joint angles and local scalar signals. The learned plateau swimming speeds were reported as

$\leq 1\,\mathrm{mm}^3$2

for the flagellar topology and

$\leq 1\,\mathrm{mm}^3$3

for the ameboid topology, with successful navigation through conflicting chemoattractants, vortical flows, moving targets, and constrictions (Xiong et al., 2024). In simulated blood capillaries with explicit red blood cells, shared-parameter PPO identified a forbidden regime in which Brownian motion and flow overwhelm propulsion, while the best success probability—about $\leq 1\,\mathrm{mm}^3$4—occurred at robot radius $\leq 1\,\mathrm{mm}^3$5 and speed $\leq 1\,\mathrm{mm}^3$6 body lengths per second (Drotleff et al., 23 Jun 2026).

Not all control need be learned. Vision-based magnetic pushing shows that a geometric “guiding corridor” and two conditions—maintaining the object inside the corridor and the microrobot behind the object—are sufficient for robust autonomous transport of micro-objects and a single CHO cell. With a $\leq 1\,\mathrm{mm}^3$7 corridor at $\leq 1\,\mathrm{mm}^3$8 Hz, the mean absolute error was $\leq 1\,\mathrm{mm}^3$9, compared with $210 \times 340 \times 50\,\mu\mathrm{m}^3$0 in open loop (Sokolich et al., 9 May 2025). A plausible implication is that smartlet autonomy can be achieved either by statistical learning, by field scheduling, or by exploiting strong geometric priors that suppress failure modes.

6. Applications, constraints, and prospective directions

The dominant application axis in the cited literature is targeted intervention in biomedical or microstructured environments. Magnetic colloidal smartlets are explicitly motivated as ultrasound-visible, navigable microrobotic swarms for biomedical environments, with an example trajectory speed of $210 \times 340 \times 50\,\mu\mathrm{m}^3$1 at $210 \times 340 \times 50\,\mu\mathrm{m}^3$2 and $210 \times 340 \times 50\,\mu\mathrm{m}^3$3 (Wang et al., 2018). COF smartlets are developed for intraocular theranostics, combining visible-to-red-light propulsion, pH-responsive release, photoacoustic imaging, OCT tracking, and hyperthermia (Sridhar et al., 2023). The clinically ready magnetic microrobot system integrates a dual Navion electromagnetic navigation system, a release catheter, and a dissolvable capsule; it achieved in-flow steering success above $210 \times 340 \times 50\,\mu\mathrm{m}^3$4 up to $210 \times 340 \times 50\,\mu\mathrm{m}^3$5, ACA targeting in $210 \times 340 \times 50\,\mu\mathrm{m}^3$6 ms, and MCA branch targeting in $210 \times 340 \times 50\,\mu\mathrm{m}^3$7–$210 \times 340 \times 50\,\mu\mathrm{m}^3$8 ms in patient-specific vascular models (Landers et al., 20 Jan 2025). MRI-powered capsules similarly couple MRI navigation with HIFU-controlled on-demand release, traveling at up to $210 \times 340 \times 50\,\mu\mathrm{m}^3$9 in ex vivo porcine small intestine and releasing drug to multiple target sites in a single operation (Tiryaki et al., 2023).

A second application axis is microassembly and modular construction. Nematic smartlets assemble one-dimensional colloidal lattices, seven-particle chains, and anisotropic patterns near wavy walls through defect-mediated transport and release (Yao et al., 2022). Micro-origami cube smartlets self-assemble at the air–water interface and via patterned hydrophobic/hydrophilic face chemistries into multi-module structures such as letters and half-registered assemblies (Lee et al., 2024). The Jasmine lineage adds an open-hardware perspective in which low-cost, replicable units perform collective perception, aggregation, communication streets, docking, and energy homeostasis (Kernbach, 2011).

The transport problem itself has also been reframed at swarm scale. In tactic run–tumble swarms, the ensemble-averaged entrainment velocity is

$(\mathrm{Fe}_3\mathrm{O}_4)$00

while tracer transport efficiency

$(\mathrm{Fe}_3\mathrm{O}_4)$01

is maximal not at perfect alignment but at intermediate directedness, with the reported optimum near $(\mathrm{Fe}_3\mathrm{O}_4)$02 (Liu et al., 29 Sep 2025). This directly challenges the intuition that stronger guidance is always better. A plausible implication is that future smartlet swarms may intentionally incorporate controlled stochasticity to maximize delivery efficiency.

The limitations are correspondingly diverse. Ultrasound visibility of magnetic colloidal smartlets depends on attaining sufficiently high area density, and single nanoparticles remain below acoustic resolution (Wang et al., 2018). UV- and $(\mathrm{Fe}_3\mathrm{O}_4)$03-driven Janus smartlets face obvious cytotoxicity constraints until visible or near-infrared catalysts and benign fuels are substituted (Bailey et al., 2021). COF smartlets must manage aggregation, especially for negatively charged, irregular TpAzo-COF (Sridhar et al., 2023). Electronic smartlets are presently memory-limited to roughly $(\mathrm{Fe}_3\mathrm{O}_4)$04 bits and translate only at about $(\mathrm{Fe}_3\mathrm{O}_4)$05–$(\mathrm{Fe}_3\mathrm{O}_4)$06 (Lassiter et al., 29 Mar 2025). Modular cube smartlets currently communicate optically only over sub-centimeter distances and operate under tight power budgets of about $(\mathrm{Fe}_3\mathrm{O}_4)$07 under one sun (Lee et al., 2024). RL capillary navigation remains demonstrated in a $(\mathrm{Fe}_3\mathrm{O}_4)$08D simulated network with flow scaled to $(\mathrm{Fe}_3\mathrm{O}_4)$09, so transfer to physiological $(\mathrm{Fe}_3\mathrm{O}_4)$10D pulsatile microvasculature remains open (Drotleff et al., 23 Jun 2026).

Prospective directions are consistent across otherwise dissimilar platforms. The magnetic ultrasound lineage points to $(\mathrm{Fe}_3\mathrm{O}_4)$11D imaging, phase-synchronized closed-loop control, and phase tagging of multiple swarms (Wang et al., 2018). The modular electronic lineage aims at smaller-node CMOS, denser heterogeneous integration, and richer sensor-programmed locomotion (Lassiter et al., 29 Mar 2025, Bandari et al., 24 Aug 2025). The physically intelligent lineage suggests networks of smartlets that cooperatively write and erase defect-mediated “circuits” or exploit embodied dynamics as an implicit sensing channel (Yao et al., 2022, Paul et al., 25 Aug 2025). Taken together, the literature indicates that microrobotic smartlets are best understood not as one device class but as a convergent microsystems program: compressing sensing, actuation, information processing, and environment-specific physical intelligence into units small enough to assemble, steer, compute, and intervene where conventional robots cannot.