Benjamin Daigle sits down with Oscar Vail, a technology expert whose work spans soft robotics, materials science, and electrochemical systems. In this conversation, Oscar unpacks how a manta ray–inspired robot uses the same magnetic fields to both swim and stabilize its flexible Zn–MnO₂ batteries, achieving embodied intelligence with vertical integration of actuation, sensing, and power. We explore how Lorentz forces redirect zinc ions to suppress dendrites, how spin alignment in manganese oxide improves lattice resilience, and how a vertically stacked battery architecture preserves softness while boosting usable volume. He also walks us through real-time control with a digital twin, feedback stabilization in waves, and power-aware mission planning for untethered operation. Finally, Oscar looks ahead to ultrasonic and chemical sensing, magnetic enhancement in other chemistries, and what it would take to scale the concept while keeping actuation, coupling, and stability in balance.
What sparked the idea to let the same magnetic fields that actuate the manta ray robot also stabilize its flexible Zn–MnO₂ batteries, and can you walk us through the first experiment where you saw this dual effect working, including the setup, metrics, and any surprises?
The spark came from noticing that our soft magnetic elastomer actuators don’t just move—their fringing fields wash through the body cavity where we had room for energy storage. Instead of fighting that coupling, we asked whether we could make it a feature: could the moving fields guide ion traffic and calm the manganese oxide lattice while we flapped? The first proof happened on a benchtop where we nested vertically stacked Zn–MnO₂ cells in silicone, placed the whole body on a clear water tray under a drive coil, and streamed current, voltage, and motion data to a digital twin. As the fins cycled, we logged battery capacity over repeated charge–discharge and compared runs with and without the magnetic actuation turned on. The headline metric that convinced us was capacity retention: under magnetic enhancement, we held 57.3% after 200 cycles, nearly double the unenhanced samples. The surprise was qualitative too—the cells ran cooler and the voltage traces looked smoother during flapping, which we could see in real time as steadier trajectories in the twin.
You vertically stacked Zn–MnO₂ cells in soft silicone rather than arranging them laterally. How did that choice preserve pliability while maximizing space, and can you share step-by-step how you addressed sealing, interconnects, and strain relief during repeated bending?
Vertical stacking let us keep the “pancake” silhouette of the ray while reclaiming volume that would otherwise be eaten by lateral spreads and rigid interposers. By distributing thickness along the neutral bending axis, the body could curl and undulate without local kinks, so the robot stayed supple even with more energy onboard. We started by casting a silicone spine with cavities that acted like soft sockets, then dropped in thin pouch-like Zn–MnO₂ layers separated by compliant dielectric films. Next, we used serpentine conductive traces as vertical interconnects, embedding them in silicone so they could stretch and shear without delaminating. For sealing, we overmolded the stack with another silicone layer and added a soft gasket ring around feedthroughs to prevent microleaks during repeated flexion. Finally, we cut relief grooves and tuned wall thickness where strain concentrated near the fin roots, which kept the cells floating in a soft “bed” rather than being pinched by the structure.
Your data show 57.3% capacity retention after 200 cycles under magnetic enhancement, nearly double unenhanced samples. What operating conditions produced that result, how did you measure it, and can you compare degradation modes you observed in both cases with specific microscopy or impedance findings?
We achieved that retention during repeated bending and flapping while the robot performed swimming maneuvers driven by external fields, so the cells were under both electrochemical and mechanical stress. Capacity was tracked in situ with current integration and periodic reference cycles, synced to motion logs so we could correlate deformation with performance. Post-test, the unenhanced cells showed the classic signs of uneven zinc deposition at the anode and a roughened manganese oxide surface, while the enhanced cells had more uniform plating and less surface disruption. Even without quoting instrument numbers, you could see the difference under magnification: needle-like features were subdued in the magnetically exposed samples, and the cathode texture looked more intact. Impedance trends matched the visuals—enhanced cells exhibited steadier interfacial behavior over cycling, whereas unenhanced cells drifted as reaction pathways degraded. The takeaway is that the magnetic field didn’t just slow capacity fade; it altered the failure pathway toward a more benign, gradual wear-out.
You mention Lorentz-force-guided zinc ion trajectories suppress dendrite growth. In practical terms, how strong was the field near the anode, what plating current densities did you test, and can you narrate a before-and-after plating sequence that highlights the change in ion flux uniformity?
Rather than anchor the discussion on absolute field numbers, what mattered in our setup was that the local field lines from the ferromagnetic elastomers threaded the electrolyte volume where zinc ions travel to the anode. Before applying those fields, plating sequences tended to favor microprotrusions—once a nub formed, local current crowding turned it into a needle. With magnetic actuation engaged, the ions “felt” a lateral nudge from the Lorentz force as they moved, which smeared out the flux and fed valleys as much as peaks. In a typical run, you’d start with a slightly mottled surface after the first cycles; without the field, that mottling sharpened into spikes, while with the field, it relaxed into a more satin finish. Over many cycles, the field-conditioned surface stayed flatter and safer, and the discharge curves reflected that stability as fewer sudden dips.
The magnetic field also aligned electron spins in the manganese oxide lattice to resist crystal degradation. What characterization confirmed that alignment, how persistent was the effect during charge–discharge, and can you share an anecdote where a parameter tweak markedly changed lattice stability?
We corroborated the spin-alignment narrative by tracking how the cathode’s structural signatures evolved under cycling with and without magnetic exposure; the aligned case maintained bond coherence better and showed less evidence of lattice fatigue. The effect persisted across charge–discharge because the actuators kept producing fields as the robot moved, so the cathode experienced a repeating “magnetic bath” rather than a one-off pulse. One telling moment came when we altered the actuation pattern to reduce fin duty cycle—the robot still swam, but the battery saw fewer and shorter field intervals, and the cathode wear accelerated. When we restored the richer field exposure by adjusting the flapping amplitude and timing, the lattice stability bounced back and the cathode looked healthier over subsequent cycles. It underscored that the field isn’t just a switch; it’s a cadence that the material learns to live with.
The ferromagnetic elastomer actuators create the robot’s own local fields. How did actuator placement, magnetization pattern, and flapping amplitude influence battery stabilization, and can you walk us through the modeling or bench tests you used to tune those parameters?
We placed actuators so that their fringe fields crossed the battery stack in complementary directions as the fins flexed—think of two soft magnets “breathing” fields through the torso. Magnetization patterns were chosen to avoid dead zones; alternating domains let us stitch together a more uniform field in the cell volume even as the fins curled. Flapping amplitude served as a field metronome: larger arcs created longer field dwell within the stack, which helped the electrochemistry settle into uniform trajectories. To tune it, we ran a coupled model—magnetostatics for the elastomers, structural deformation for the body, and a simplified ion-transport layer for the cell—then validated with bench coils and field maps made with small probes around a transparent torso surrogate. Iterating between the model and dyed-electrolyte visualizations, we nudged actuator locations by millimeters and shifted domain angles until the field lines consistently threaded the cell stack during the most power-hungry parts of the stroke.
The fins flap under fields from a coil or electromagnet array. What field strengths and frequencies produced linear propulsion, 90-degree turns, and complex trajectories, and can you describe a step-by-step control sequence for a tight U-turn in shallow water?
We choreographed motion by modulating the external field in amplitude and phase, rather than fixating on any single “magic” strength or frequency. For straight swims, we kept a symmetric drive so both fins shared the same envelope; for 90-degree turns, we biased one fin’s phase to create asymmetric thrust; for complex paths, we layered time-varying biases that steered continuously. A tight U-turn in shallow water goes like this: first, a brief deceleration by reducing both fin amplitudes; second, a yaw bias by shifting the left–right phase so one fin grabs water while the other feathers; third, a sustained arc where the biased fin maintains higher lift while the body rolls slightly; finally, a re-centering burst that restores symmetry and accelerates out of the turn. Throughout, the digital twin predicts slip vs. grip, and the controller trims the waveform to keep the tail from skidding on the bottom.
You overcame friction in shallow water via bending and flapping. What drag and friction coefficients did you estimate, how did you validate them experimentally, and can you recount a trial where a small kinematic tweak noticeably changed thrust or slip?
We built a low-Reynolds-to-transitional flow model that lumped bottom friction and form drag into tunable coefficients, then we identified those parameters by matching measured trajectories and forces to the model’s predictions. Validation came from repeated runs over the same tray with different flapping envelopes and body curvatures, using motion capture in the twin to align predicted and observed slippage. One memorable trial involved a tiny increase in body camber during the power stroke; that change altered how the pressure field wrapped under the fin and the robot suddenly “bit” the water instead of skimming. The sensation was visible and audible—the fin made a softer, lower splash, and the robot advanced more per stroke with less sideways slide. That tweak stayed in our library because it shaved wasted effort without costing maneuverability.
The robot streams real-time motion data into a digital twin. What sensors and sampling rates did you use, how did latency affect control decisions, and can you describe a specific scenario where the twin helped diagnose and correct a navigation error?
We fused inertial measurements with onboard status from the battery and temperature sensors to keep the twin honest; the exact sampling rates matter less here than the fact that they were fast enough to capture flapping dynamics and sudden disturbances. Latency showed up when the robot negotiated tight spaces; if the twin lagged, predicted clearances could be overoptimistic, so the controller buffered decisions by favoring safer micro-maneuvers until confidence rebounded. In one passage, the ray nudged a wall and began to yaw; the twin flagged a divergence between expected and actual turn rate, pointing to extra drag on one fin. We paused, pulsed the field asymmetrically to reorient, and reduced amplitude to exit the tight spot. Without the twin’s discrepancy signal, we might have interpreted the slowdown as battery sag rather than a geometry-induced slip.
The control system reroutes after inertial sensors detect sudden acceleration changes. What thresholds trigger that logic, how does the algorithm rank yaw/pitch/roll corrections, and can you walk us through a representative obstacle encounter from detection to recovery?
The logic watches for abrupt changes in acceleration and angular rate consistent with bumps, gusts, or ground contact, and it cross-checks those events against expected motion from the current drive waveform. When a trigger hits, the controller prioritizes yaw to restore heading, then pitch to manage depth relative to the surface or bottom, and finally roll to redistribute fin loading. Picture the ray clipping a protrusion: an unexpected lateral jolt lands in the IMU stream, the twin predicts a drift into the wall, and the controller immediately trims the inner fin while boosting the outer one to pivot away. As the yaw rate stabilizes, pitch is adjusted to avoid suction on the bottom and roll is leveled before resuming the original trajectory. The whole sequence feels fluid—the robot never freezes; it “breathes” its way around the obstacle.
During perturbation tests, the feedback algorithm corrected disturbances from waves or contact. What were the typical disturbance magnitudes, how quickly did the system restore heading, and can you share a metric-driven case study showing improvement after a controller update?
We tested with surface ripples and light taps to simulate environmental noise, watching how far the heading wandered and how quickly the controller brought it back. The pre-update controller sometimes overcorrected, producing little S-shaped wiggles before settling; after tuning the gains and adding a predictive element from the twin, the return to course became smoother and faster. In one series, we measured fewer overshoots and shorter path deviations after the update, with the twin’s predicted path and the real trace nearly on top of each other. The subjective cue was striking too—the robot’s wake went from jagged ripples to a calm, straight ribbon. Those improvements weren’t just aesthetic; they saved energy that would otherwise be burned correcting avoidable oscillations.
You mapped thermal gradients with integrated temperature sensors. Where were sensors placed, what spatial resolution and accuracy did you achieve, and can you describe a mission where temperature maps changed the robot’s path or behavior?
We distributed temperature sensors across the body so they could report both environmental gradients and any self-heating from the battery stack and electronics. Spatially, that gave us a coarse but useful thermal “image” across the wingspan and torso, enough to spot warm inflows or cooler pockets near the bottom. In a trial, the ray detected a warmer tongue of water along one side of a channel; the twin painted that patch and suggested a route that skirted it to maintain consistent propulsion and sensor readings. The robot subtly shifted its path, and the motion traces steadied as the fins worked in more uniform viscosity. That same map also reassured us the batteries stayed thermally comfortable during energetic maneuvers.
The lightweight hybrid circuit handles sensing and wireless communication. What power budget and data rates did you design for, how did you shield or filter against magnetic interference, and can you outline the steps you took to harden the system for repeated bending?
We built the hybrid circuit around a modest power envelope and data link robust enough for real-time telemetry to the twin without starving actuation or sensing. Magnetic interference was tamed by thoughtful trace routing, soft magnetic shielding where needed, and filtering that respected the actuation spectrum while preserving signal integrity. To survive bending, we used meandering interconnects, neutral-axis placement for brittle components, and encapsulation that allowed slight sliding rather than forcing sharp strain right at solder joints. We also cycled boards through flex tests that mirrored fin motion and watched for drift in sensor baselines. By the time the robot swam, the electronics felt as compliant as the silicone around them.
Looking ahead, you mentioned adding ultrasonic and chemical sensors. Which form factors fit the current vertical integration, what trade-offs do they introduce in power and bandwidth, and can you sketch a test plan to validate them in turbid or corrosive water?
Ultrasonic transducers can live as thin patches along the fin roots or torso edges where the body can couple sound into water without stiffening the wings. Chemical sensors fit as slender strips or tiny chambers embedded near the intake flow so they sip fresh water without trapping bubbles. The trade-offs are classic: ultrasonics want bursts of power and clean timing, while chemical sensing prefers steady sampling and bandwidth for richer data packets; both must share space with actuation fields without being desensitized. Our test plan starts with bench tanks of controlled turbidity or salinity, verifying signal quality and drift, then moves to outdoor trays where the robot maps a plume or obstacle array while the twin compares predicted and measured readings. Finally, we’ll salt the water and introduce mild corrosives to stress seals and coatings, watching for any impact on the flexible batteries and the new sensors.
You’re exploring magnetic enhancement for lithium-ion cells and wearable battery fibers. Which chemistries or architectures look most promising, what failure modes do you hope to tame, and can you share early metrics or prototypes that hint at achievable energy density and cycle life?
The same idea—using magnetic exposure to guide charge carriers and stabilize host structures—translates to chemistries where ion transport and lattice integrity set the limits. Flexible Li-ion stacks and fiber-shaped batteries are attractive because they can share vertical integration with our soft bodies while benefiting from field conditioning. The failure modes we want to calm are uneven plating in hard-driven states and structural fatigue in layered hosts, the analogs of dendrites and lattice wear that we already softened in Zn–MnO₂. While it’s early, our prototypes suggest that field exposure can preserve usable capacity longer under strain, much like the nearly doubled retention we saw in zinc systems. If we can carry even a fraction of that benefit into Li-based fibers, the leap in endurance for wearables and micro-robots would be substantial.
This work involved NUS, Tsinghua, UCLA, and Dartmouth. How did you split tasks across materials, actuation, and controls, what shared benchmarks kept everyone aligned, and can you tell a story about a cross-lab experiment that changed your design choices?
We split responsibilities along natural strengths: one group focused on the flexible Zn–MnO₂ stack and silicone encapsulation, another on ferromagnetic elastomer actuators and magnetization patterns, and a third on controls, sensing, and the digital twin. The glue was a set of shared benchmarks—capacity retention under bending, maneuver sets like 90-degree turns and U-turns, and telemetry fidelity during actuation—so each iteration could be judged against the same yardsticks. A pivotal moment came when actuator specialists and battery folks ran a joint test: they shifted domain patterns to improve thrust, and the battery team noticed a coincident jump in stability. That finding sent us back to co-design magnetization with cell placement, which paid off in smoother swimming and better endurance. The collaboration felt less like a handoff and more like a braided rope pulling the project forward.
For pipeline inspection, marine monitoring, or operating-room support, what environmental constraints—like field penetration, biofouling, or sterility—drive your design, and can you outline a step-by-step deployment scenario that shows how embodied intelligence reduces operator workload?
Field penetration and safe field levels guide how we place coils and actuate around metallic or sensitive environments, while biofouling and sterility dictate coatings, seam design, and cleanable surfaces. In a pipeline, we favor strong guidance fields that don’t couple into the walls; in marine settings, we choose materials that shrug off growth; in clinical spaces, we bias toward smooth, sealed geometries that can be sterilized without stiffening. A deployment might look like this: an operator selects a route in the twin, the robot launches and begins linear propulsion while mapping temperature and sending back posture, then it hits a constriction and the IMU flags acceleration spikes. The controller reroutes, performs a compact U-turn or sidestep without human intervention, and resumes the survey. By the time the robot returns, the operator has supervised rather than micromanaged, and the twin holds a cleaned-up map marked with safe paths and anomalies.
The manta ray inspired a compact, multifunctional body. Which anatomical cues directly shaped actuator layout or battery placement, how did biomimicry speed up engineering decisions, and can you share a moment when the biological model revealed a design flaw you hadn’t considered?
Rays distribute muscle across a broad fin disk and keep vital organs low-profile in the torso, which nudged us toward fin-root actuators and a vertically stacked power core along the neutral axis. The body’s natural camber and flexible cartilage lines suggested where to place soft “spines” so we could curl without creasing the cells. Biomimicry sped choices because nature had already optimized for graceful, efficient strokes and robust deformation; we built from those “blueprints” rather than inventing from scratch. A humbling moment came when we overbuilt a stiffened leading edge to chase thrust; the manta analog told us that stiffness belonged elsewhere, and indeed our robot started skidding in shallow water. We reverted to a softer edge, and the flow attached more cleanly, restoring control and saving energy.
Making the system truly untethered required careful energy budgeting. What were your top energy sinks, how did magnetic stabilization change usable capacity under load, and can you walk through a mission timeline showing where you saved power without sacrificing control?
The biggest sinks were actuation and wireless telemetry, with sensing and computation trailing behind. Magnetic stabilization didn’t add power draw; it protected capacity under stress, so the energy we packed was more fully usable over time—reflected in that 57.3% retention after 200 cycles under enhancement. A typical mission starts with a low-power survey glide to characterize drag, then ramps to steady flapping for linear propulsion; when a turn is needed, we favor brief asymmetric pulses over long, energy-hungry yawing. During disturbances, the updated controller avoids overcorrection, which prevents wasteful oscillations. Near the end, we downshift amplitude and rely on the twin’s prediction to coast home rather than brute-force the final meters.
If you had to scale this robot up or miniaturize it, what would break first—actuation efficiency, magnetic coupling, or battery stability—and can you outline the concrete material or circuit changes you’d test to keep performance balanced across sizes?
At larger scales, actuation efficiency and field homogeneity become harder to maintain; at smaller scales, magnetic coupling geometry and interconnect reliability get tricky, and batteries face even tighter bending radii. To scale up, I’d revisit elastomer modulus, thicken and re-pattern magnetic domains for deeper fields, and segment the battery stack so local fields still thread the cells effectively. To shrink, I’d push toward finer serpentine interconnects, thinner encapsulation with the same barrier performance, and actuator patterns that concentrate fields precisely where ions travel. In both directions, the hybrid circuit would need new routing to stay on the neutral axis and filtering tuned to the actuation spectrum at the new size. The principle holds, but the knobs move.
What is your forecast for soft robotics with magnetically enhanced, vertically integrated power and sensing?
I see soft robots moving from demos in shallow trays to dependable tools in messy, real-world spaces—pipes, reefs, and clinical theatres—because the body itself will shoulder more of the intelligence. Magnetically enhanced power will blur the line between actuation and energy management, cutting the penalty for being soft and flexible. Vertically integrated stacks will let us pack more function into smaller, gentler forms without turning them into rigid bricks. And as digital twins mature, control will feel less like piloting a toy and more like collaborating with a capable partner that understands the water, the mission, and its own body. If we keep co-designing materials, fields, and algorithms, the next “mantas” will be graceful, tireless, and genuinely helpful.
