For decades, glaciologists have worked with a hypothesis known as marine ice cliff instability, or MICI. The idea is elegantly simple: an ice cliff taller than roughly 100 meters — about the height of the Statue of Liberty — cannot support its own weight and should collapse. When it does, it exposes an even taller cliff further inland, which should then collapse in turn, triggering a cascading retreat that progressively dismantles the ice sheet behind it. Previous studies built this picture mainly from physical models of ice mechanics, scaled up to cliff dimensions and anchored by only a handful of real-world observations.

Researchers at the Massachusetts Institute of Technology and the Woods Hole Oceanographic Institution decided to test that assumption at scale. Graduate student Caroline Needell, together with Catherine Walker of WHOI and Jeremy Bassis of the University of Michigan, analyzed satellite measurements of more than 15,000 ice cliffs around Antarctica, tracked over four years from 2019 to 2023. Their data came from NASA’s Ice, Cloud, and Land Elevation Satellite-2, or ICESat-2, which sweeps over the polar regions and uses a laser altimeter to measure the height of ice sheets, glaciers, sea ice, and the surrounding ocean surface. Quarterly measurements allowed the team to watch individual cliffs grow, collapse, and recover through time.

The first surprise was height. The proposed critical threshold for cliff collapse sits at about 100 meters, yet the average Antarctic ice cliff in the dataset measured only about 30 meters tall. Cliffs exceeding the theoretical limit turned out to be rare. That finding alone suggests the preconditions for runaway collapse are less widespread than models had implied, because most of the ice front simply never reaches the height at which instability is supposed to begin.

But the team did not stop at static snapshots. They zoomed in on the four regions where ICESat-2 recorded the tallest cliffs, each roughly 100 meters high, including one that stood 127 meters above the water. Tracking these giants every three months over four years, they found that the tallest cliffs did collapse, and the ice sheet behind them retreated — but at a pace faster than MICI models had predicted. Rapid loss of ice, in isolation, would translate into greater contributions to sea level rise than anticipated.

The second surprise was what happened next. The retreat did not cascade. Instead, it was short-lived, and the glaciers in every region eventually stabilized, with ice spreading back out toward the ocean. “We found that when the cliffs were retreating, they were doing so faster than what has been modeled, which would mean more ice loss and higher contributions of sea level rise,” Needell said. “But because we found that the retreat was short-lived, and all the glaciers eventually stabilized, that would suggest that things won’t be so bad. We don’t see this runaway effect.”

What stops the cascade? The study points to stabilizing mechanisms that height-only models ignore. One is mélange — the slushy mixture of broken icebergs and sea ice that packs together against the front of a cliff. The researchers measured mélange height and found that taller cliffs are backed by thicker mélange, which appears to act as a built-in buttress, propping up the ice wall and absorbing the stress that would otherwise drive further calving. Cliffs that held their position over time were consistently those with the most substantial mélange cover.

Water depth plays a role as well. The satellite data showed that the height of the surrounding ocean can stabilize a cliff: the deeper the water, the more it buoys and pushes back against the ice face, resisting collapse. In effect, the ocean itself can act as a support structure, a counterintuitive result given that deep water is usually associated with faster glacier retreat through undercutting and melting.

Beneath the ice, another safeguard emerged. The shape of the land under the ice sheet — the bed topography — can function as a literal speed bump. Where the underlying terrain is flat, the slope of the overlying ice reflects its true thickness. But where the bed rises, the ice sitting over that bump is thinner than it appears from the surface. Thinner ice means shorter, more stable cliffs, less likely to fail. When the researchers compared cliff behavior with topographic measurements of Antarctica, they found that ice cliffs indeed stabilized around regions where the land rose upward. The team also observed that the taller a cliff is, the faster it thins out: as collapsing cliffs settle into shorter, more stable configurations, the process naturally arrests itself rather than accelerating.

“It’s this rapid retreat, punctuated by periods of stabilization,” Needell said, describing the pattern that initially puzzled the team. The resolution, she explained, was to consider everything that affects cliff height rather than height alone. “Now we know what to look for,” she said. “We really need to know the bed topography, and we need to know the ice thinning rate, in order to get ice cliff instability right, because this value really matters in the context of projecting sea level rise.” The message for modelers is clear: predictive ice-sheet models that treat cliff height as the only variable — triggering runaway retreat once the 100-meter threshold is crossed — are missing the feedbacks that nature apparently provides. The work was supported in part by NASA, the U.S. National Science Foundation, and the DOMINOS project, a component of the International Thwaites Glacier Collaboration. By grounding MICI theory in more than 15,000 real cliffs observed over four years, the study replaces a frightening but poorly tested hypothesis with a more nuanced picture: Antarctic ice cliffs can fail dramatically and quickly, but the ice sheet behind them has more ways to catch its breath than scientists once believed.

Subject of Research: Marine ice cliff instability and the stability of Antarctic ice sheet margins observed by satellite altimetry

Article Title: Satellite data reveal how Antarctica’s ice cliffs behave over time

Article References: Satellite data reveal how Antarctica’s ice cliffs behave over time. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: Antarctica, ice cliffs, marine ice cliff instability, ICESat-2, sea level rise, ice shelves, glaciology, mélange, bed topography, MIT, Woods Hole Oceanographic Institution, Science

Researchers in the laboratory of Professor Kripa Varanasi at MIT, working with collaborators at MIT and Johns Hopkins University, have developed a handheld device capable of gently collecting living cells from precise locations within tissue samples. Described in the journal Device, the system is designed to sample the fallopian tubes, where roughly two decades of research have revealed that many cases of high-grade serous ovarian cancer, the most common form of the disease, actually begin. Because precursor lesions in the fallopian tubes can be microscopic and extraordinarily difficult to sample, they have remained largely invisible to conventional diagnostic workflows.

The technical concept behind the device is elegantly simple, even if its execution required careful engineering. A 3D-printed microfluidic chip is pressed against the surface of freshly excised tissue, where one syringe creates a vacuum that seals the device against the specimen, ensuring that nothing leaks during operation. A second syringe then pushes fluid through a small microfluidic channel, producing a flow that applies shear stress parallel to the tissue surface. Rather than cutting or scraping cells away with a scalpel or brush, the flowing liquid gently detaches living cells from tiny, targeted sections of tissue, leaving the surrounding structure intact for conventional pathological examination.

“We came up with this device where one syringe creates a vacuum that holds it against the tissue, and a second syringe pushes liquid through it,” says Bert Vandereydt, an MIT postdoc and co-first author of the study. “The vacuum creates a seal, so nothing leaks, and then we locally apply what is basically a microfluidic chip on the tissue that gently shears the cells off.” The insight emerged from the lab’s deep experience with fluid-surface interfaces, and from conversations with surgeons at Johns Hopkins who observed that certain cells were loose enough to come off during routine washing and other procedures in the operating room.

The significance of collecting cells alive, rather than fixed in preservative, cannot be overstated. The traditional approach to examining removed fallopian tubes involves placing the tissue in a chemical preservative that maintains structure but kills the cells, making it impossible to grow them in culture. A pathologist must then painstakingly search thin sections under a microscope for cancerous or precancerous regions, a process that Varanasi describes as time-consuming and destructive to the cells. With the new technique, researchers can not only see what cells look like but recover them viable, cultivate them, grow them into organoids, and build living models of disease that show how cells from a specific patient behave.

In their study, the researchers demonstrated that the shear stress applied by the device can be tuned to different levels, and they compared their approach against other cell-collection workflows. The cells harvested with the fluidic method remained viable and grew in culture far more readily than those detached by conventional means. The tunability extends across disease types as well. Loose prostate cancer cells detach under stress as low as one pascal, whereas bone cancer cells proved far more tenaciously adherent, with only a few cells releasing even under five pascals of applied stress. As Vandereydt puts it, the approach is agnostic to the disease, which suggests broad utility well beyond ovarian cancer.

Proving the technology in real clinical conditions required an unusual level of dedication. The team tested the device on fresh human fallopian tube samples shipped from collaborators at Johns Hopkins, which meant staying on call for deliveries that could arrive at unpredictable times. “Sometimes, we’d get an email from our collaborators at 11 p.m. saying ‘There are two fallopian tubes coming tomorrow,'” Vandereydt recalls. After each experiment, the samples were returned for conventional pathology, ensuring that nothing was lost from the diagnostic record. From those fresh specimens, the team successfully collected living cells and converted them into organoids, the crucial building blocks for testing, disease modeling, and eventually personalized treatment development.

The project traces its origins to a fortunate convergence of clinical and engineering expertise. Rebecca Stone, a gynecologic oncologist at the Johns Hopkins University School of Medicine, has long advocated for risk-reducing fallopian tube removal for certain women at elevated risk, a procedure that can be performed after childbearing years while largely preserving normal hormone production. When Angela Belcher, MIT’s James Mason Crafts Professor, introduced Stone to Varanasi, the collaboration was born, funded by Break Through Cancer, a foundation that assembles interdisciplinary teams to attack the most stubborn problems in oncology. “This is exactly the kind of problem that benefits from bringing clinicians and engineers together,” Stone says. “We understand the clinical need, while the MIT team brings a very different perspective from fluid mechanics and engineering.”

The road ahead involves two distinct phases. In the near term, the researchers plan to apply the device to tissue that has already been removed from the body, a pathway that offers easier regulatory approval and could transform routine histopathology by allowing living cells to be harvested while the specimen is preserved for standard examination. Stone notes that integrating the technology into existing pathology workflows could create a powerful new way to study carcinogenesis and fundamental biology directly from human tissue. In parallel, Belcher’s group is developing optical approaches to identify suspicious regions of tissue, and the cell-collection device could then extract living cells from exactly those flagged locations. The longer-term ambition is bolder still: adapting the device to swab samples inside patients directly, enabling easier testing and earlier detection without surgery.

What makes this work resonate far beyond a single laboratory is the vision it points toward for cancer medicine as a whole. If clinicians can harvest living cells from specific, suspicious regions without destroying them, then diagnostics could evolve from static snapshots of preserved tissue into dynamic studies of living disease. Organoids grown from a patient’s own cells could be exposed to different therapies in the laboratory, revealing which treatments work before a patient receives a single dose, a genuine step toward personalized medicine. Varanasi credits Break Through Cancer with making the collaboration possible, noting that teams working on pancreatic cancer, brain cancer, leukemia, and other diseases all expressed how valuable better methods for obtaining living cells from specific tissue regions would be. For the researchers, the ultimate measure of success is human: “If this work can ultimately help women by enabling earlier detection of ovarian cancer, I would find that incredibly fulfilling,” Varanasi says. “That is really what motivates us — taking the science and engineering we develop in the lab and using it to make a difference in people’s lives.”

Subject of Research: A handheld microfluidic shear-stress device for collecting viable cells from fallopian tube tissue to improve early cancer detection

Article Title: New cell-collection device could improve early cancer detection

Article References: New cell-collection device could improve early cancer detection. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: ovarian cancer, microfluidics, fallopian tubes, early detection, organoids, living cells, personalized medicine, MIT, shear stress, cancer diagnostics, pathology, biomedical engineering

The stakes of extending forecast range are not abstract. In February 2021, three severe winter storms struck Texas, triggering the worst energy infrastructure failure in the state’s history. At least 246 people died as a result of the cascading disruptions, which cut off food, water, heat, and electricity across wide swaths of the state. According to the researchers, if energy officials had received reliable warnings weeks in advance, they could have taken concrete mitigating steps, such as arranging additional trucks of natural gas to serve disrupted areas before the crisis unfolded. The Texas disaster is a vivid illustration of why the subseasonal window, spanning roughly two to six weeks ahead, is considered a key planning horizon for emergency response, agriculture, and the energy sector.

To understand why forecasts decay so quickly, the researchers point to a foundational insight from MIT’s own history. In the 1970s, MIT Professor Edward Lorenz developed chaos theory, describing how infinitesimally small differences in the atmosphere’s initial conditions compound into completely different weather outcomes. Lorenz proposed a theoretical two-week limit beyond which deterministic weather prediction becomes impossible, because the noise in the system grows until it overwhelms any initial signal. That limit, says Judah Cohen, an MIT research scientist and co-author of the new study, exerted a strong influence on the field and may have discouraged researchers from even attempting forecasts beyond two weeks. Cohen, however, never fully accepted the barrier. While the atmosphere can in principle produce an infinite number of patterns, he argues, it tends to fall into favorite patterns repeatedly, so that once one pattern emerges, the sequence of subsequent dominoes becomes, to some degree, predictable.

Today’s best forecasts come from two main families of models. Dynamical models numerically simulate the physical evolution of the atmosphere and oceans over time, while artificial intelligence models learn to make predictions from vast archives of historical weather data. AI models have recently shown impressive gains at short ranges, but at longer time horizons they perform even worse than dynamical models. The gap between these regimes is what forecasters call the subseasonal range, and Cohen describes it bluntly as the predictability desert. In the short range, forecasters can exploit the initial conditions of the climate system and atmosphere; on seasonal and longer scales, slow-varying boundary forces such as El Niño provide leverage. The subseasonal range, caught in between, can take advantage of neither, and in most cases, Cohen says, forecasting there has been no different from rolling dice at a casino.

The breakthrough came from an idea credited to Cohen’s longtime collaborator Lester Mackey, a researcher at Microsoft: rather than trying to build a better forecasting model from scratch, use machine learning to quantify and remove the errors of existing ones. The team built what they describe as a wrapper, or postprocessing layer, that sits on top of an existing forecasting system. Unlike the more simplistic error-correction schemes used by other forecasting systems, the new layer uses AI to learn from a forecasting system’s historical mistakes and shifts its forecasts accordingly. Because the target range demands probabilistic forecasts, in which the likelihood of a range of outcomes is given rather than a single deterministic prediction, the correction operates on distributions of possibilities. Crucially, the model is adaptive: as more forecasts and observational data become available, it continuously refines its correction.

The researchers trained and developed the model using MIT Engaging, a supercomputing cluster hosted by MIT’s Office of Research Computing and Data, but the finished model does not require a supercomputer to run, a practical detail that lowers the barrier for meteorological services worldwide. To validate the approach, the team last year submitted three models incorporating their postprocessing layer to the European Center for Medium-Range Weather Forecasts, widely regarded as the best forecasting center in the world. ECMWF holds forecasting competitions between models each season, providing a rigorous, independent benchmark. The winning submission combined ECMWF’s dynamical forecasting model with another model and then applied the postprocessing correction on top. The result was a doubling of the original model’s predictive power in the subseasonal range and, for the first time, a demonstrated improvement over simply relying on long-term climatological averages.

That last achievement may be the most consequential. Until now, Cohen notes, no one had shown a model that predicts weather better than long-term averages for subseasonal forecasts; previous systems had essentially all been worse, to the point that flipping a coin could outperform them, and researchers had questioned whether investing in models for this range was worthwhile at all. The team’s models consistently beat climatology. Equally important, the approach produced a statistically significant improvement in predicting extreme weather events, such as flooding and cold spells. This is a notoriously difficult target. When ECMWF developed its own non-AI error correction, the corrected model actually got worse at predicting extremes. Cohen attributes that failure to the nature of traditional bias correction, which always pulls forecasts toward the mean, whereas extreme weather sits on the fringe of the probability distribution. The AI-driven correction, by contrast, learned to preserve and sharpen signals at the edges.

The technique’s versatility is another of its strengths. In the Nature paper, the team showed that the method could improve land forecasts across most regions of the globe. They also applied the bias-correction wrapper to other forecasting models and found it improved those as well, lifting even lesser models to parity with ECMWF’s forecasts. Cohen emphasizes that the model is an add-on wrapper rather than a replacement: the team used it with ECMWF’s dynamical model because that model was considered the best, but it can be applied to any model, and it even improves models that have already been bias-corrected using another AI algorithm. The improvement is robust, and the tool is open source and low cost, making it accessible to national weather services that lack the computational resources of the major forecasting centers.

Real-world adoption may already be beginning. Cohen says the meteorological office for the Republic of Kenya has reached out about implementing the model to improve forecasting of the annual monsoon season, with the goal of boosting agricultural yields. The potential applications extend across sectors: farmers could time fertilization around expected rain, protect crops from incoming freezes during winter harvests, and take steps to avoid drought losses. Energy and emergency management officials stand to benefit as well. Cohen recounts conversations with energy officials who explain that two days of warning before an ice storm is not enough time to prepare, but two weeks could be sufficient to make adequate arrangements, from securing fuel supplies to staging repair crews. The work was supported in part by the U.S. National Science Foundation, the Carlos Chagas Filho Foundation for Research Support, and the European Union, and the researchers hope that other groups around the world will build on the openly available model to push the predictability desert further back.

Subject of Research: Machine learning postprocessing to improve subseasonal weather forecasting accuracy from two to six weeks ahead

Article Title: Researchers improve weather forecasts looking two to six weeks out

Article References: Researchers improve weather forecasts looking two to six weeks out. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: subseasonal forecasting, machine learning, weather prediction, MIT, ECMWF, error correction, extreme weather, chaos theory, Nature, energy planning, emergency preparedness, climatology

The technique targets a long-standing bottleneck in cellular immunotherapy. Approved CAR T cell therapies, seven of which have been cleared by the U.S. Food and Drug Administration over the past ten years, all follow the same demanding recipe. Doctors draw blood from a patient, ship the sample to a specialized manufacturing facility, isolate the T cells, engineer them to express a chimeric antigen receptor, a synthetic protein that recognizes a specific target antigen, expand the engineered cells to sufficient numbers, and then infuse them back into the patient. Each step adds time, cost and logistical fragility, and the same challenges have carried over into efforts to build therapies around natural killer cells, a different branch of the immune system with its own distinct advantages.

Natural killer cells are innate immune cells that specialize in detecting and destroying stressed, infected or malignant cells, and they can be armed with chimeric antigen receptors just as T cells can. For one class of diseases, they hold a decisive edge over engineered T cells. When the therapeutic target is a T cell itself, whether a malignant T cell in lymphoma or an autoreactive T cell driving an autoimmune condition, CAR T cells run into a fundamental problem: T cells engineered to recognize T cell proteins can attack one another, depleting the therapy and undermining the treatment. Natural killer cells face no such fratricidal conflict, because they are a different cell type and can eliminate dysfunctional T cells without destroying their own kind.

NK cells can treat these T cell malignancies, not only T cell cancer but also T cell autoimmune diseases, the researchers note. Yet manufacturing large quantities of CAR NK cells outside the body poses many of the same difficulties as generating CAR T cells, and no CAR NK therapy has yet been approved by the FDA, although many groups are working toward that goal. The MIT team set out to remove the ex vivo manufacturing step entirely by generating CAR NK cells in vivo, inside the patient’s own body, without removing any cells. If a patient can simply receive an infusion of nanoparticles, the process would be dramatically simplified and the costs substantially reduced, the researchers say.

The delivery vehicle at the heart of the study is the lipid nanoparticle, the same class of fatty carrier that made mRNA vaccines possible. In this application, the particles encapsulate strands of messenger RNA encoding a chimeric antigen receptor, the instructions a cell needs to build the targeting protein on its own surface. Because the payload is mRNA rather than DNA, the reprogramming is transient: the NK cells produce the receptor for a limited period, which offers a built-in safety margin and allows dosing to be adjusted. The researchers designed the particles to carry receptors directed against specific antigens, including CD19, a protein found on B cells, and CD3, a protein found on T cells, including the autoreactive ones implicated in autoimmune hepatitis.

Getting the particles to the right cells is the other half of the engineering problem. Lipid nanoparticles administered systemically tend to be taken up by liver cells and other scavenging tissues rather than by circulating immune cells. To solve this, the MIT team coated the particles with antibody fragments that bind to receptors on the surface of NK cells, ensuring that the payload is delivered preferentially to the intended targets. When the particles were infused into mice, the researchers found that they successfully reprogrammed roughly 80 percent of the targeted NK cells, a strikingly high efficiency for an in vivo cell-engineering approach and one that suggests a meaningful fraction of a patient’s native NK population could be conscripted without any laboratory manipulation.

The therapeutic results in animal models spanned both halves of the team’s ambition: cancer and autoimmunity. In a mouse model of an aggressive form of melanoma, three of the five mice that received nanoparticles reprogramming their NK cells to target the cancer cells went into complete remission. In a humanized mouse model, one in which mice receive transfusions of human blood cells and transiently generate a human-like immune system, the researchers used the particles to treat T cell lymphoma. After five injections of the reprogramming nanoparticles, tumors were eradicated, and the outcome was comparable to what the team achieved with conventionally manufactured CAR NK cells, the kind produced by removing NK cells, engineering them in the lab and returning them to the animals.

The autoimmune experiments may prove the more novel contribution, since they demonstrate that in vivo reprogramming can be aimed at a self-destructive immune response rather than a tumor. Autoimmune hepatitis occurs when autoreactive T cells attack hepatocytes, the functional cells of the liver, progressively damaging the organ. When mice with the disease received nanoparticles encoding a receptor directed against those T cells, the autoreactive population was destroyed and the liver damage healed, with these effects observed after a single injection. The researchers hope engineered NK cells of this kind could eventually be applied to a broader range of autoimmune disorders driven by dysfunctional T cells, naming multiple sclerosis and rheumatoid arthritis as potential targets.

The work is an example of the power of nano delivery systems and their potential to treat cancer and other diseases, according to Robert Langer, the David H. Koch Institute Professor at MIT and a co-author of the study. The research was led by senior author Daniel Anderson, a professor in MIT’s Department of Chemical Engineering and a member of the Koch Institute for Integrative Cancer Research and the Institute for Medical Engineering and Science, together with lead authors Yuang Chen, Julia Monkovic and Haseeb Mughal. The study was funded by Sanofi, the Marble Center for Cancer Nanomedicine, and a Koch Institute Support Grant from the National Cancer Institute. Much work remains before the approach can be tested in patients, from confirming safety and dosing to demonstrating that the nanoparticles can efficiently reach NK cells in humans as they do in mice. But if the in vivo strategy translates, it would replace a weeks-long, facility-intensive manufacturing pipeline with a ready-to-infuse product, potentially widening access to engineered cell therapy far beyond the specialized centers that can currently offer it.

Subject of Research: In vivo reprogramming of natural killer cells with ligand-directed lipid nanoparticles for cancer and autoimmune disease immunotherapy

Article Title: Reprogrammed NK immune cells target cancer and autoimmune disorders

Article References: Reprogrammed NK immune cells target cancer and autoimmune disorders. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: natural killer cells, CAR NK cells, lipid nanoparticles, messenger RNA, immunotherapy, T cell lymphoma, autoimmune hepatitis, melanoma, MIT, Nature Biotechnology, cell therapy, nanomedicine

The robot, which the researchers call an aquabot, is built around a skeleton made from a film of gel roughly the length and width of a stick of chewing gum. The two halves of the gel serve as the robot’s fins, and each fin is coated with a layer of live skeletal muscle cells far thinner than a single strand of human hair. These cells are genetically engineered to contract when illuminated, which means the robot can be steered entirely with light. When the team shines light on one fin, the muscle cells on its surface twitch in unison, causing the whole fin to flap with enough force to pull the robot through the water. By flashing light on one fin or the other, and by varying the timing of those flashes, the researchers can control both the direction and the speed of the swimming robot.

In demonstrations, the paper-thin bot swam and swiveled its way through a simple watery maze set inside a large petri dish, following a light source that the researchers maneuvered manually over the surface of the water. At its fastest, the aquabot covered a distance of about four times its body length in one minute. That pace is admittedly modest compared with elite human swimmers, who can cover up to sixty-five body lengths per minute, but it is competitive with more leisurely swimmers in the animal kingdom. The cow shark, for instance, explores the ocean at roughly the same rate relative to its body size. For a robot whose entire propulsion system is a film of muscle cells thinner than a hair, the performance is a striking proof of concept.

It takes a lot of force to move through water versus air, notes study author Ritu Raman, associate professor of mechanical engineering at MIT. The robot’s quite strong, given its size, she says. That strength did not come by accident. It was the product of a deliberate engineering campaign to optimize every parameter of the platform on which the living muscle grows, a process the researchers describe as tuning the robot’s skeleton. The lesson of the work is that in biohybrid robotics, the artificial scaffold matters just as much as the biological engine that sits on top of it.

The new aquabot builds directly on a milestone the same group achieved the previous year. In that earlier effort, Raman’s laboratory created an iris-inspired disk of artificial muscle tissue by stamping a disk of gel with a pattern of concentric and radial grooves and then depositing live muscle cells onto the gel’s surface. The cells grew along the grooves, forming a thin layer, and when stimulated with light they moved in patterns that stretched and squeezed the disk much as a human iris dilates and constricts the pupil of the eye. That work was the first to show that muscle cells could be grown in a very thin layer and in complex patterns that, when stimulated, could move in multiple controllable directions. People hadn’t seen this muscle architecture engineered from scratch before, Raman says, and the cells were moving in multiple directions. But they only moved about one hundred microns. From a robotics perspective, their movements were tiny.

The goal of the new study was to close that gap: to maximize muscle movement so that the tissue could produce enough force to do useful mechanical work, such as powering a swimming robot. The key, the team discovered, was the underlying gel. In the iris-inspired design, the muscle cells had been grown on fibrin, an ultrasoft gel that the researchers realized could quickly shrivel in response to the forces generated by the contracting muscles. A scaffold that deforms under load wastes the very force the cells are trying to deliver. To better support the cells and maximize their output, the engineers focused on tuning three properties of the gel: its composition, its stiffness, and the size and shape of the grooves stamped into its surface. For engineering any type of tissue, it’s known that these are knobs you can tune, Raman says. And we wanted to optimize all these parameters to support live muscle cells.

The optimization process was systematic. The team experimented with multiple formulations of gel at different stiffnesses, each stamped with grooves of different geometries. One groove type resembled a skinny square trough, while another was more of a long curved valley. When muscle cells were deposited onto each type of grooved gel, the researchers found that the cells settled into alignment most effectively in grooves that were more square than curved. This matters because alignment is everything in engineered muscle: cells that line up in the same direction tend to fuse into fibers, and those fibers form a stronger, more coordinated muscle tissue capable of generating useful force in a single direction. Square grooves, the team concluded, were the way to go.

Composition and stiffness proved equally consequential. Instead of fibrin, the researchers switched to gelatin methacrylate, or GelMA, a material widely used in tissue engineering. By preparing different recipes of GelMA, they created skeletons of varying stiffness and observed how muscle cells grew on each. The cells grew in better alignment and produced the most force on stiffer gels, a finding that runs counter to the intuition that softer scaffolds are always gentler on living tissue. The team also varied the thickness of the gel and found that a film just half a millimeter thin offered good support for a single layer of muscle cells while remaining light enough that the cells stayed anchored to the gel when they contracted, rather than peeling away. As a final step, the researchers exercised the muscles, subjecting them to a training routine of flashing lights that strengthened the tissue over time, much as workouts build strength in an athlete.

With the pumped-up cells and the optimized gel in hand, the team assembled a thin, two-finned robot: the gel, stamped on both sides with square-bottomed grooves and lined with muscle cells that fused into fibers and eventually formed a strong, aligned muscle tissue. You can think of the robot as having two independent muscles, Raman says. If we shine a light on just one, only that muscle moves. If shining on both, they both flap. This independent control of each fin is what gives the aquabot its maneuverability, allowing it to turn and navigate obstacles rather than simply drifting forward in a straight line.

The researchers acknowledge that the current design is relatively basic in form; their first aim was simply to show that the bot could produce enough force to swim at all. Their next goal is to optimize the body design to enable faster swimming. But even at slow swim speeds, the possibilities are intriguing. One could imagine a muscle-powered swimmer being used for purposes like environmental monitoring in aquatic environments, Raman says. Biohybrid robots powered by living muscle could one day perform delicate jobs like exploring environments too fragile or unpredictable for conventional hardware, because living tissue is soft, responsive to its surroundings, and can heal itself. The approach also promises practical advantages over existing biohybrid designs, which are typically built from bulky, three-dimensional chunks of lab-grown skeletal muscle that require millions of cells to fabricate. Thinner, less bulky designs like the aquabot could be cheaper to build and could move more efficiently, bringing soft, light-driven, self-healing machines a step closer to real-world duty beneath the water’s surface.

Subject of Research: Biohybrid robotics using engineered skeletal muscle thin films to power light-controlled swimming robots

Article Title: Powered by muscle cells, a paper-thin robot swims through watery maze

Article References: Powered by muscle cells, a paper-thin robot swims through watery maze. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: biohybrid robots, muscle cells, soft robotics, aquabot, MIT, light-activated actuators, GelMA, tissue engineering, swimming robot, skeletal muscle, environmental monitoring, Advanced Functional Materials

Lipid nanoparticles are the workhorses of a rapidly growing class of therapeutics. mRNA injected on its own would be shredded within minutes by enzymes in the body, so it must be wrapped in a protective shell of lipids that both shields the genetic cargo and encourages cells to take it up. These are, as the researchers describe them, a revolutionary type of medicine, but one that depends entirely on a delivery vehicle capable of reaching the right cells. Getting that vehicle right has long been as much art as science.

A typical LNP contains four ingredients: an ionizable lipid that binds the genetic payload, a phospholipid, cholesterol, and a PEG-lipid that stabilizes the particle. To assemble them, manufacturers rapidly mix two streams of fluid — one carrying lipids dissolved in ethanol, the other carrying mRNA in an acidic buffer. Conventional mixing uses roughly three times more of the mRNA solution than the lipid solution, a ratio that promotes particle formation but offers little command over the dimensions of the resulting particles. Because size governs biodistribution, that blind spot has been a persistent bottleneck.

Size matters enormously. A particle destined for the liver may fail entirely if it is too large or too small to be taken up by the intended tissue, and tuning dimensions can open the door to targeting other organs. As senior author Allan Myerson, a professor of the practice in MIT’s Department of Chemical Engineering, notes, the problem may look simple at first glance but in reality requires a deep understanding of how lipid nanoparticles assemble. Previous production methods, he explains, simply could not reliably control particle size and shape.

The foundation for the new platform was laid in a study the same team published last year, also in ACS Nano. There, the researchers split the mixing process into two steps. In the first, mRNA and lipids are combined at equal flow rates, allowing particles to begin nucleating and growing. Then, after a precisely timed delay, additional buffer is injected, which halts growth. By adjusting the residence time — the interval between the two mixers — the team gained a direct lever on particle size. Long delays let particles grow large; short delays keep them small. Changing the composition of the buffer added in the second step, meanwhile, allowed them to morph the particles from spheres into elongated forms resembling avocados.

Crucially, both interventions adjust size and shape without altering the chemical composition of the nanoparticles themselves, meaning developers can explore the design space of a fixed formulation. Cedric Devos, an MIT postdoc and one of the lead authors of the study, compares the residence time to a lever over LNP manufacturing that simply was not available before. The approach gives formulators a way to ask, systematically, which combinations of parameters generate specific size and shape attributes before testing which particles perform best in biological experiments — a powerful development tool, in his words.

The new paper takes that methodology a decisive step further: automation. The team connected their two-step mixing system to a commercially available dynamic light scattering device that measures particle sizes in real time as they form. The entire process runs without any human intervention. A researcher simply specifies a target size, and the system produces particles, measures them, and — if they miss the mark — automatically adjusts the delay time and other operating parameters until the output matches the specification. The platform can also generate particles of different shapes, though shape measurements currently must be performed outside the automated loop.

With the first study, the team unlocked a new methodology for truly engineering lipid nanoparticles by size and shape, says Peter Sagmeister, an MIT postdoc and lead author. With the second, they automated the whole process. The project also drew on MIT’s Undergraduate Research Opportunities Program, through which applied mathematics and computer science undergraduates Joy Ren, Sofiya Chubich, and Dylan Nguyen helped integrate advanced software engineering with chemical engineering to build the automated platform — a collaboration Sagmeister credits with giving the students hands-on experience at the intersection of the two disciplines.

The payoff of automation is speed, and speed enables something deeper: data. Working on a timescale far faster than conventional experiments allow, the researchers systematically varied the inputs to the system and recorded how particle size and shape responded. They then used this dataset to train a machine-learning model capable of predicting the combination of factors needed to generate a particular size or shape on demand. In effect, the platform converts LNP formulation from a craft into an engineering discipline, where desired properties can be specified up front and achieved predictably rather than discovered through weeks of iterative tinkering.

The implications reach across the entire pipeline of nucleic acid drug development. Because a 150-nanometer particle and a 70-nanometer particle built from identical ingredients will behave very differently in the body, as Devos points out, the ability to rapidly generate matched particles differing only in size gives developers a rigorous way to identify the optimal formulation for a given application — whether a vaccine, a cancer therapy, or a treatment aimed at a tissue that current LNPs struggle to reach. The researchers have filed for a patent on the technology and are commercializing it through a startup called BIZON Labs, which has received support from MIT’s Researcher 2 Entrepreneur program and acceptance into the Institute’s flagship delta v accelerator. The work was funded by the U.S. Food and Drug Administration and a Koch Institute Support grant from the National Cancer Institute, and carried out in part using MIT.nano facilities. If the platform performs as promised outside the lab, the slow, guesswork-laden search for the perfect nanoparticle may soon give way to something resembling an assembly line for the next generation of genetic medicines.

Subject of Research: Automated two-step microfluidic manufacturing and machine-learning optimization of lipid nanoparticles for RNA and DNA therapeutics

Article Title: A new technique could accelerate the development of RNA therapies

Article References: A new technique could accelerate the development of RNA therapies. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: lipid nanoparticles, RNA therapeutics, mRNA vaccines, microfluidics, drug delivery, machine learning, automation, nanotechnology, chemical engineering, MIT, ACS Nano, particle size control

The system, called SANDO — short for Safe AutoNomous trajectory planning for Dynamic unknOwn environments — was developed by Kota Kondo, who recently completed his doctorate in aeronautics and astronautics at MIT, together with Jesús Tordesillas of Comillas Pontifical University in Madrid, MIT graduate students Juan Rached, Lili Sun, and Yixuan Jia, and senior author Jonathan P. How, the Ford Professor of Aeronautics and Astronautics at MIT. The work appears in the journal IEEE Transactions on Robotics. Its central claim is striking in its simplicity: the only prior knowledge the planner requires is the maximum speed an obstacle could possibly reach. Given that single number, the algorithm can chart a course through a completely unmapped, dynamic environment and prove, before the drone ever takes off, that the resulting trajectory will remain free of collisions.

“In the hardest possible environment, where the UAV has no map of the area and there are unknown obstacles moving around, we established a mathematical guarantee of safety,” Kondo says. “The only thing the planner needs to know is the top speed the obstacles could reach. Given that, you could use it in any environment, without a map, and you know the UAV is not going to crash into anything.” That guarantee matters because formal safety proofs have long been the missing ingredient in autonomous flight. Trajectory planners typically rely on images and data streamed from a drone’s onboard cameras and sensors to chart a path toward a goal, and most existing approaches fall into one of two camps. Some are designed for unknown but static environments, where obstacles never move. Others handle dynamic obstacles loosely, steering around them heuristically without any formal proof that the robot will not crash. Computing every possible collision scenario in a dynamic world, meanwhile, would take far too long to be useful in real deployments.

“In an unknown dynamic environment, you don’t have many assumptions to rely on,” Kondo explains. “In those types of environments, researchers haven’t yet been able to mathematically guarantee that a trajectory is safe.” The MIT team’s answer was to build rigor into the mathematics from the ground up. Rather than treating safety as a property to be checked after a path is generated, they designed an algorithm whose output is provably safe by construction, covering every possible case, including the awkward edge cases that often break real-world robotic systems. The result is a planner that does not merely hope to avoid obstacles but can demonstrate, mathematically, that it will.

The key to SANDO’s guarantee lies in how it structures the space around the drone. The system begins by mapping out what the researchers call a safety corridor: a series of connected regions of three-dimensional space that the robot can travel through, each of which is guaranteed to contain no obstacles. Safety corridors are not new in themselves, but conventional versions suffer from a fundamental weakness in dynamic settings. A corridor that is clear at one moment can be invaded seconds later by a moving obstacle, silently invalidating the safety proof. SANDO addresses this by making the corridor time-sensitive. Instead of treating obstacles as fixed points, the system explicitly considers their possible future movements when constructing the corridor, so that the guaranteed-clear regions remain valid as the world changes around them.

To make those future predictions possible without knowing exactly where an obstacle will go, SANDO employs a dedicated module that detects, groups, and monitors dynamic obstacles in the environment. For each moving object it identifies, the module uses the obstacle’s maximum velocity to compute how far it could possibly travel within a given timespan. It then encloses the obstacle in a sphere that captures the farthest distance it could reach in any direction over that interval. Even without knowing the obstacle’s actual trajectory, the drone knows the obstacle must remain inside that sphere. SANDO then builds its safety corridor around these spheres, ensuring the planned path can never intersect a region a moving obstacle could occupy. “In the real world, obstacles are going to move, so the safety corridor you create at one point won’t be useful as things move into the corridor,” Kondo says. “But because we consider this time component, we can now guarantee safety into the future.”

Safety alone, however, is not enough if the resulting path is slow or wasteful. To keep flights efficient, SANDO adds a second layer of intelligence: a heat-map-based planner that identifies the “hot” regions of the environment — areas crowded with obstacles — and steers the drone away from them. By biasing the search away from danger zones before the fine-grained trajectory optimization begins, the system charts a more direct and efficient course around hazardous areas rather than threading through them. Once a collision-free safety corridor has been established, SANDO optimizes the trajectory within that corridor to find the fastest possible path to the goal, all while respecting the corridor’s boundaries.

The optimization itself is where much of the engineering effort went. As the robot travels, SANDO continuously adjusts the safety corridor and reformulates the trajectory, ensuring the path remains collision-free all the way to the destination. To make this feasible in real time, the researchers developed a set of mathematical tricks that simplify the optimization problem, allowing the drone’s onboard computer to rapidly recalculate trajectories and react to sudden changes in the environment without relying on offboard processing. The payoff of that rigor, Kondo notes, came in flight testing. “The most difficult part of developing SANDO was the math,” he says. “When you try to guarantee safety, you need to be rigorous and ensure your theory covers every possible case, even edge cases. Once we had that mathematical guarantee, it was very easy to fly the UAVs.”

The experimental results support the theory. In simulation, SANDO reached the robot’s goal faster than several state-of-the-art planning systems while completely avoiding collisions in every environment tested. The team then moved from simulation to the physical world, flying a real UAV equipped with SANDO across twelve test flights. In every one of those flights, the system avoided all dynamic obstacles it encountered, using only the robot’s onboard computer and sensors to replan safe trajectories on the fly. For a field in which simulated guarantees frequently dissolve on contact with real sensors, wind, and latency, that record of zero collisions across both simulated and physical trials is a notable demonstration that the mathematical proof translates into practice.

The potential applications extend well beyond the laboratory. A drone carrying SANDO could fly deep into a wildfire, dodging falling branches and sudden flare-ups while gathering critical information for rescuers on the ground. It could search collapsed buildings for survivors, explore networks of hidden mine tunnels, or deliver packages across a crowded neighborhood where pedestrians, vehicles, and other drones share the airspace. The researchers see several directions for future work: making the system even more computationally efficient, and combining it with machine-learning models that would let users issue instructions to the robot in plain language rather than through technical interfaces. The research was funded in part by the Defense Science and Technology Agency of Singapore. For now, SANDO stands as a rare example of a promise kept in robotics: a planner whose safety is not a hope or a statistic, but a theorem — one that holds even when the world ahead is entirely unknown and everything in it is moving.

Subject of Research: Guaranteed collision-free autonomous trajectory planning for drones in dynamic unknown environments

Article Title: Planning system ensures a robot’s flight path will remain collision-free

Article References: Planning system ensures a robot’s flight path will remain collision-free. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: UAV, trajectory planning, MIT, SANDO, collision avoidance, autonomous navigation, safety corridor, dynamic obstacles, robotics, IEEE Transactions on Robotics, search and rescue, motion planning

The award places particular emphasis on the central thread running through Acemoglu’s career: the analysis of how political institutions condition economic institutions and, through them, the long-term prosperity of nations. This framework, developed over more than three decades of research, has become a point of reference for economists and political scientists seeking to explain why some countries grow rich while others remain trapped in poverty. Rather than treating geography, culture, or natural resources as the decisive factors, Acemoglu’s work argues that the rules governing political and economic life, who holds power, how that power is constrained, and whether opportunities are broadly shared, are fundamental determinants of national success and failure.

The jury’s citation connected this scholarship directly to the present moment. At a time when fundamental democratic principles, including the rule of law, the protection of fundamental rights, and respect for pluralism, face new challenges in different parts of the world, the jury stated, his work provides important theoretical insights and valuable analytical elements for informed public debate and the design of public policies. The statement underscores a distinctive feature of Acemoglu’s research: its explicit engagement with questions of institutional design and democratic resilience, areas that sit at the intersection of technical economics and urgent contemporary politics.

Acemoglu’s path to prominence began with an academic formation spanning two countries. He graduated from the University of York in the United Kingdom in 1989 and completed his doctorate at the London School of Economics in 1992, remaining at the institution for a year as a teacher. In 1993, at a remarkably young age for such an appointment, he joined the faculty of the Massachusetts Institute of Technology, where he has built his career ever since. From that base he has produced a body of work whose scale is unusual even by the standards of elite economics departments.

That output includes more than a hundred articles published in the discipline’s leading international journals, among them the American Economic Review, the Quarterly Journal of Economics, and the Review of Economic Studies, as well as four books. His standing within the profession is reflected in his membership of the American Academy of Arts and Sciences, the European Economic Association, and the Econometric Society, and in his service as editor of Econometrica, widely regarded as one of the most rigorous and selective journals in economic theory. Few economists have combined such editorial and institutional responsibilities with a comparable record of published research.

Among his most impactful works is Why Nations Fail: The Origins of Power, Prosperity, and Poverty, written with James A. Robinson, a book that brought their institutional argument to a global readership far beyond academic economics. At its core lies the concept of inclusive institutions: those that encourage investment and innovation and provide a level playing field for participants in economic life. The book contrasts these with arrangements that concentrate power and opportunity in the hands of narrow elites, an imbalance that, in the authors’ account, suppresses the creative destruction and broad participation on which sustained prosperity depends. The distinction between inclusive and extractive institutions has since become a standard analytical vocabulary in development economics and comparative politics.

A second fundamental axis of Acemoglu’s research concerns technological progress and its economic and social consequences. His work has examined how innovation transforms economies and, critically, how its benefits and costs are distributed across different groups in society. This line of inquiry has become increasingly prominent as automation, artificial intelligence, and digital platforms reshape labor markets in wealthy and developing countries alike. By treating technology not as an exogenous force but as a set of choices shaped by institutions, incentives, and power, Acemoglu’s research asks who decides which technologies are developed and whose interests those decisions serve, questions that have moved to the center of public policy debates on the future of work.

The honors accumulated across his career trace the trajectory of a scholar whose influence was recognized long before the field’s highest award arrived. He received the John Bates Clark Medal from the American Economic Association in 2005, an honor traditionally reserved for economists under the age of forty who have made significant contributions to the discipline, and the BBVA Foundation Frontiers of Knowledge Award in 2016. The culminating recognition came in 2024, when he was awarded the Nobel Memorial Prize in Economic Sciences together with Simon Johnson and James A. Robinson for their studies of how institutions are formed and how they affect prosperity, a prize that confirmed the centrality of the institutional research program the three scholars built over decades of collaboration.

The medal Acemoglu will receive in December carries its own distinguished history. The CSIC Extraordinary Medal for Scientific Merit, formerly known as the CSIC Gold Medal, has recognized since its creation in 1989 individuals and entities that have made exceptional contributions to the advancement of science and technology. Its roster of past recipients reads as a cross-section of modern science: the British theoretical physicist Stephen Hawking, honored in the medal’s inaugural year of 1989; the American Nobel laureate physicist Roy J. Glauber in 2008; the Colombian neuroscientist Rodolfo Llinás in 2012; the astrophysicist Jocelyn Bell Burnell, whose discovery of pulsars transformed astronomy, in 2015; the Nobel laureate in Physics Donna Strickland in 2023; and the Spanish biophysicist Eva Nogales, recognized in 2024 for her pioneering work in visualizing macromolecular function using cryo-electron microscopy.

Acemoglu’s addition to that list is notable for a simple reason: he is an economist joining an honor roll dominated by natural scientists, a reflection of the CSIC’s judgment that his contributions to understanding institutions, democracy, and development rank among the exceptional advances in knowledge of recent decades. As democracies worldwide confront pressures on their institutions and as technological change accelerates faster than the frameworks governing it, the jury’s rationale reads as both a tribute to a career and a statement about the questions that matter now. The December ceremony will formally mark what the citation makes clear: a body of work that has given scholars, policymakers, and citizens a shared language for discussing why nations succeed, why they fail, and how the rules societies write for themselves shape the prosperity and freedom their people enjoy.

Subject of Research: Award of the CSIC Extraordinary Medal for Scientific Merit to economist Daron Acemoglu for his research on institutions and economic development

Article Title: CSIC awards its Extraordinary Medal for Scientific Merit to economist Daron Acemoglu

Article References: CSIC awards its Extraordinary Medal for Scientific Merit to economist Daron Acemoglu. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: Daron Acemoglu, CSIC, Extraordinary Medal for Scientific Merit, institutions, economic development, political economy, democracy, technological progress, Nobel Memorial Prize, Why Nations Fail, MIT, Spain

The discovery fills a curious gap in astronomers’ understanding of how stars and their companions interact. Across the galaxy, most planetary bodies circle their host stars in stable, detached orbits, like Earth around the sun. At the other extreme, astronomers have documented a handful of instances in which a star pulls a planet too close and swallows it whole — a fate that awaits Earth itself when the sun eventually swells into a red giant. ZTF J0440+2325 sits strikingly between these two extremes. Instead of a sudden engulfment or a peaceful coexistence, the system shows a star gradually feeding on its companion at a measured, sustainable pace. “When we think of stars interacting with planets or brown dwarfs, the picture is always that the star eventually swallows the other thing,” said Kevin Burdge, assistant professor of physics at MIT. “But here, we’ve found an alternative: Instead of swallowing the thing up, the star can gradually eat it, for billions of years.”

The story of the discovery began several years ago, when Burdge was combing through data from the Zwicky Transient Facility, or ZTF, a survey that uses a camera on a telescope at Palomar Observatory in California to scan the sky for rapid changes in brightness. Such flickers can signal supernovae, gamma-ray bursts, or colliding neutron stars. Light curves from supernovae typically resemble a bell curve, charting the gradual brightening and fading of an explosion. What Burdge noticed instead was something far stranger: a triangular waveform, rising and falling in a sharp, repeating pattern that no known stellar process seemed to produce. “I remember first looking at this and thinking, stars don’t make triangular waveforms like this,” he recalled.

At the time, Burdge and his colleagues were investigating a different class of objects known as black widow binaries, in which an extremely dense, rapidly spinning neutron star slowly consumes a much smaller companion star, much like the spider that gives the systems their name. Burdge wondered whether the triangular signal might come from a similar system. But the light from the source refused to cooperate. In black widow binaries, the light appears to wobble because a very light object whips around a much heavier one, and the amplitude of that wobble encodes the extreme imbalance of masses. “We weren’t seeing that whipping back and forth here,” Burdge said. “It didn’t make any sense. We couldn’t explain what this was.” The signal sat unresolved for years, a persistent puzzle in the team’s files.

The breakthrough came when Burdge and Aaron Householder, a graduate student in MIT’s Department of Earth, Atmospheric, and Planetary Sciences, decided to revisit the mystery. From the original ZTF signal, they pinpointed the source’s location within the Milky Way, about 300 light years away, and trained multiple telescopes on the object. Their measurements revealed a wobble — but a far smaller one than any black widow binary would produce. “That was the real clincher for this system,” Householder said. “When we measured that wobble, we found we were not seeing a black widow. This was a low-mass star that’s orbited by a brown dwarf. The wobble was too small in amplitude to be anything else.” The insight that unlocked the puzzle was that a system with less total mass allows two objects to orbit each other gently, without the dramatic back-and-forth whipping seen in David-and-Goliath pairings.

The measurements established that the two objects are extraordinarily close, with the brown dwarf completing an orbit every 87 minutes. Both bodies are small by stellar standards: the star weighs roughly 85 times the mass of Jupiter, while the brown dwarf comes in at about 25 Jupiter masses. With two low-mass objects circling at such intimate range, the team suspected that one might be pulling material from the other — a process known as accretion. Accretion is most often observed around objects that are extremely massive yet physically tiny, such as black holes and neutron stars. When a black hole draws material from a nearby companion, the matter spirals inward through a disk. “The difference here is, the thing absorbing matter is not a tiny black hole but a star, which is relatively big in size,” Burdge explained. “So matter just pummels directly onto the surface, at very high speeds, like an asteroid hitting the moon.”

To test whether accretion was actually occurring, the researchers ran simulations of the system. Using the measured properties of the star and the brown dwarf, they modeled particles of matter on the brown dwarf’s surface and tracked how those particles would behave over time under the laws of physics and the equations of motion. The result was unambiguous: the test particles fell directly onto the surface of the star. “When we track those test particles, we see they indeed fall right onto the surface of the star,” Householder said. “This is the first time we’ve caught a low-mass star actively accreting from another low-mass object.” The finding establishes a new category of interacting binary, one in which the accretor is a full-fledged star rather than a compact remnant.

The team also calculated the rate of the transfer, and the numbers are remarkable in both directions. The brown dwarf is losing material at about one one-hundred-thousandth of an Earth mass per year — the equivalent of roughly 40 million dump trucks’ worth of material, or about 1.3 trillion one-pound burritos, every second. Yet even at that staggering everyday scale, the loss represents only a tiny fraction of the brown dwarf’s total mass. The transfer is slow and steady enough that, given the size of the system, the star could continue its leisurely snacking for hundreds of thousands of years, and possibly for billions. What would be a cataclysmic feast on human timescales is, cosmically speaking, a grazing meal.

The slow accretion also solved the mystery of the triangular light curve. The stream of material from the brown dwarf strikes the surface of the red dwarf, heating a large, glowing hot spot at the point of impact. Viewed from afar, the system’s brightness rises and falls as the brown dwarf and its stream circle the star, carrying the blazing impact zone in and out of view. “It’s like you’ve got this continuous fireball onto one of the objects, and as one orbits the other, that hotspot comes in and out of view, and the peak of the triangle signal is when you’re looking right at the fireball,” Burdge explained. The triangular waveform, once an inexplicable anomaly, is now a diagnostic signature of a star being pummeled by a steady stream of stolen matter.

With the puzzle resolved, the team is now turning its attention to finding similar systems elsewhere in the galaxy. Because the triangular light curve is so distinctive, it offers a ready-made search template for future surveys. “It’s inspiring a lot of new searches on our part,” Householder said. “I think we’re going to learn a lot about a different kind of way that planets and brown dwarfs interact with their host stars.” The research, supported in part by the National Science Foundation, included MIT co-authors Aaron Householder, Kaitlyn Shin, Saul Rappaport, Joheen Chakraborty, and Emma Chickles, along with collaborators from Caltech, the University of Hawaii, the Instituto de Astrofísica de Canarias, the Universidad de La Laguna in Spain, and the Harvard-Smithsonian Center for Astrophysics. For now, ZTF J0440+2325 stands as proof that the drama between stars and their companions is not always a story of sudden destruction — sometimes, it is a meal savored over eons.

Subject of Research: Stable mass transfer from a brown dwarf onto a low-mass star in a close binary system

Article Title: Astronomers catch a star slowly snacking on a brown dwarf, 300 light years away

Article References: Astronomers catch a star slowly snacking on a brown dwarf, 300 light years away. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: brown dwarf, red dwarf, accretion, binary star system, ZTF J0440+2325, Zwicky Transient Facility, light curve, stellar evolution, mass transfer, MIT, Nature Astronomy, black widow binary

The research, led by senior author Navid Azizan, the Alfred H. and Jean M. Hayes Career Development Associate Professor in MIT’s Department of Mechanical Engineering and the Institute for Data, Systems, and Society, appears in the journal IEEE Transactions on Pattern Analysis and Machine Intelligence. Azizan, who is also a principal investigator in the Laboratory for Information and Decision Systems, worked with lead author Zeyang Li, a graduate student in mechanical engineering and LIDS, and Kaveh Alim, a graduate student in IDSS and LIDS. Their central insight is deceptively simple: when it comes to satisfying hard constraints, what matters is the model’s final output, not the intermediate steps along the way. Yet most existing methods do exactly the opposite, clamping the model at every step of generation and, in doing so, strangling its ability to find good solutions.

To understand why this matters, it helps to look at how pretrained generative models actually work. Models such as Stable Diffusion and FLUX learn to create new data by transforming random noise into structured output through a sequence of denoising steps. This iterative process is what gives them their creative power: they explore a rich space of possibilities before settling on an answer. When users need those answers to obey strict requirements — safety rules, physical laws, task-specific limits known as hard constraints — the standard approach is a technique called projection-based sampling. At each intermediate step, the partial solution is forcibly projected onto the set of constraint-satisfying values, dragging the sample back into the feasible region over and over again.

The problem, as the MIT team recognized, is that constraining the entire generation process is both wasteful and counterproductive. The internal trajectory of a generative model is discarded once the final output is produced, so there is no reason every intermediate sample must be feasible. Forcing feasibility at every step removes degrees of freedom the model could otherwise use to explore, and it can prevent the sampler from ever reaching a better final solution. Worse, projection-based methods typically focus solely on constraint satisfaction, ignoring the opportunity to improve other qualities of the answer — for instance, shortening a robot’s trajectory while keeping it collision-free. An answer that merely avoids crashing is not the same as a good answer.

HardFlow flips the logic. Instead of enforcing constraints at every intermediate step, the algorithm gives the model freedom to roam during generation and enforces the hard constraints only on the final output. The trick is in how it steers the process. The researchers reformulated hard-constrained sampling as a trajectory-optimization problem, borrowing tools from optimal control theory. This framing treats the sequence of denoising steps as a controlled trajectory that can be nudged toward a goal, with subtle corrections applied along the way and the constraints imposed at the destination. Control theory, Azizan notes, provides a powerful framework for formalizing the optimal way to make those corrections — turning what might otherwise be ad hoc clamping into a principled steering problem.

Of course, formulating the problem is easier than solving it. The trajectory-optimization problem wraps around an enormous neural network, potentially hundreds of interconnected layers deep, and optimizing over such a structure directly would be computationally intractable at deployment time. The MIT team’s solution was to exploit the mathematical structure of flow-matching models, which allowed them to decompose the global trajectory problem into a sequence of smaller, single-step subproblems. Through systematic transformations and carefully chosen approximations, they derived an efficient, scalable algorithm that preserves the key properties of the original optimization problem while remaining fast enough to run when the model is actually being used. Essentially, as Azizan describes it, the team transformed the trajectory-optimization problem into something that can be solved very efficiently at deployment time.

This deployment-time operation is one of HardFlow’s most practical advantages. The technique is plug-and-play: it can be applied to any pretrained generative model without retraining, fine-tuning, or modifying the network’s weights. In an era when powerful foundation models are widely available and constantly being updated, a method that layers constraint satisfaction on top of existing models — rather than baking it in through expensive retraining — dramatically lowers the barrier to deploying generative AI in safety-critical contexts. Hospitals, factories, and autonomous systems do not need to rebuild their models; they simply need a smarter sampler.

The reformulation as an optimization problem also buys something projection methods cannot offer: the ability to jointly optimize for quality. Because HardFlow treats constrained generation as an optimization, additional objectives can be folded directly into the framework. A robotic path planner using HardFlow, for example, does not merely find any collision-free route from one machine to another — it can find the shortest or quickest collision-free route. Li emphasizes that this joint handling of feasibility and quality is what allows the framework to perform substantially better than existing methods, which typically treat constraint satisfaction as the sole goal and leave solution quality to chance.

The experimental evidence spans three domains: robotic manipulation, maze navigation, and text-guided image editing. Across all of them, HardFlow achieved perfect constraint satisfaction — every single output met the required constraints — while consistently outperforming baseline methods on measures of solution quality. In the robotics experiments, the algorithm enabled a robotic manipulator to avoid collisions with obstacles while simultaneously finding the quickest path to a target object. Most competing methods, by contrast, either produced collisions outright or found paths that took significantly more time to execute. Notably, HardFlow’s computational cost was comparable to or lower than that of most rival techniques, meaning the gains in quality and safety did not come at the price of impractical runtimes.

The implications reach well beyond the lab. Generative models are increasingly being proposed for roles in which their outputs must obey nonnegotiable rules: robots sharing space with human co-workers, controllers governing physical infrastructure, and vision systems whose errors carry real consequences. HardFlow offers a way to preserve the exploratory power that makes generative AI valuable while guaranteeing that the final answers respect the boundaries the real world imposes. As Azizan puts it, the promise of generative AI lies in its ability to explore a rich space of possibilities, but the real world places limits on which possibilities are acceptable — and his team’s approach is designed to preserve that generative power while enforcing the nonnegotiable requirements of high-stakes applications. The researchers note that future work could extend the framework to settings where the AI model itself is also updated during use, allowing constraint satisfaction and sample quality to improve in a more adaptive, closed-loop manner. For now, HardFlow stands as a demonstration that the path to trustworthy generative AI may not run through bigger models or more training data, but through smarter control of the generation process itself.

Subject of Research: Hard-constrained sampling for pretrained generative AI models via trajectory optimization

Article Title: New method enables AI for safety-critical situations

Article References: New method enables AI for safety-critical situations. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: HardFlow, generative AI, diffusion models, flow-matching models, hard constraints, trajectory optimization, optimal control, robotics, constraint satisfaction, MIT, safety-critical AI, sampling algorithms