Scientists have successfully translated an entire viral genome into a format a quantum computer can process and analyze — the first time this has been accomplished at full-genome scale.

The work, which involved encoding the genetic sequence of a virus as quantum data, was published and reported by researchers who say the approach could eventually allow quantum systems to identify viral mutations, model genomic interactions, and assist in drug discovery at speeds classical computers cannot match.

What the researchers did

The team converted the nucleotide sequence of a complete viral genome — the ordered string of adenine, thymine, guanine, and cytosine base pairs that encodes all viral instructions — into qubits, the basic unit of quantum information. Unlike classical bits, which hold a value of either 0 or 1, qubits exploit superposition to represent both states simultaneously, allowing a quantum processor to evaluate many possible sequence configurations at once.

Translating a genome for quantum analysis requires mapping each nucleotide or group of nucleotides to a corresponding quantum state. The process is non-trivial: genomic data is inherently sequential and high-dimensional, while current quantum hardware operates with limited qubit counts and high error rates.

Getting a complete viral genome — even a relatively compact one — to fit within those constraints required data compression and encoding strategies tailored to the quantum architecture.

The viral genome selected for this experiment is small relative to bacterial or human genomes, but the encoding itself represents a meaningful proof of concept for how biological sequence data can interface with quantum processing systems.

Why quantum hardware is being applied to genomics

Classical genomic analysis tools — including BLAST for sequence alignment and machine learning models trained on genomic databases — are computationally intensive. Comparing a novel pathogen’s genome against a reference library, or predicting how mutations affect protein folding, demands significant processing time and energy when done at scale.

Quantum computing offers a theoretical speed advantage for certain classes of problems, including optimization and pattern matching across large combinatorial spaces.

Genomic sequence analysis, which often involves searching for the best alignment among millions of possibilities, fits that profile. Quantum technology is also being explored for improvements in semiconductor physics, with researchers finding that quantum phenomena in solid-state materials may open new device architectures.

For genomics specifically, the potential application is not just speed. Quantum algorithms such as the Harrow-Hassidim-Lloyd (HHL) algorithm are designed to solve linear systems exponentially faster than classical counterparts, which could transform how researchers model gene expression networks or viral replication dynamics.

Hardware constraints and open questions

Current quantum computers operate in the noisy intermediate-scale quantum (NISQ) era — devices with dozens to a few hundred physical qubits that are prone to decoherence errors. Encoding an entire genome does not mean a NISQ device can analyze it with the same fidelity as a fault-tolerant quantum computer would. The present achievement is closer to a compatibility demonstration than a functional genomic analysis pipeline.

Error correction remains a central obstacle. Physical qubits must be grouped into logical qubits with redundancy to suppress noise, which multiplies the hardware requirement substantially.

For reference, a truly fault-tolerant quantum computer capable of running complex genomic algorithms at scale could require millions of physical qubits — far beyond today’s hardware. The research is framed as an early but concrete step toward quantum-enabled biology, not an operational system.

There is also the question of data input bandwidth. Loading large biological datasets into a quantum system — a process called quantum random access memory (QRAM) — remains an active area of research with no widely agreed-upon hardware implementation.

Path to practical application

Translating a viral genome for quantum processing establishes that the conceptual pipeline — from biological sequence to quantum representation — is viable at the genome scale. The next engineering challenge is running meaningful quantum algorithms on that data and benchmarking their output against classical tools.

Pandemic preparedness is one area where faster genomic analysis would have concrete value. During the COVID-19 pandemic, tracking variant emergence required continuous re-sequencing and comparison of viral genomes across global databases.

A quantum-accelerated version of that workflow could, in principle, compress the timeline from sequence detection to variant characterization. Whether current or near-term quantum hardware can deliver that in practice remains an open question that this research alone cannot answer.

Researchers operating a deep-sea submersible have found the world’s largest known whale graveyard on the ocean floor, uncovering dozens of whale carcasses and a thriving ecosystem of creatures — many believed to be new to science — feeding off the remains.

What the submersible found

The discovery was made during a submarine dive that brought cameras and sensors to depths where light does not penetrate. Researchers observed multiple whale carcasses scattered across the seafloor in close proximity, forming what scientists are now calling the largest concentration of whale fall sites ever recorded in a single location.

Whale falls — the term for a whale carcass that sinks to the seafloor after death — are known to sustain entire biological communities for decades. Soft tissue attracts mobile scavengers first, followed by collagen-consuming organisms, and finally chemosynthetic bacteria that break down lipids embedded in the bones. Each stage can support a distinct faunal assemblage, and the full succession can span 50 to 100 years on a single carcass.

At this site, researchers reported seeing animals that they believe had not been previously documented by science. The density of carcasses in one location is particularly unusual; most documented whale falls are isolated, stumbled upon by chance during unrelated deep-sea surveys.

The biology of a whale fall

A whale carcass functions as a nutrient island in an otherwise food-scarce environment. In the deep ocean, organic material from the surface is sparse by the time it drifts to abyssal depths.

A single large whale can deliver tens of thousands of pounds of organic carbon to the seafloor in a single event — a concentrated pulse of energy that would otherwise take decades to accumulate through marine snow, the slow drift of organic particles from surface waters.

Chemosynthetic bacteria, analogous in ecological function to those found at hydrothermal vents, colonize whale bones during the final decomposition stage. These bacteria oxidize hydrogen sulfide produced by the anaerobic breakdown of bone lipids, generating energy without sunlight.

This process can sustain specialized invertebrates — including polychaete worms, clams, and gastropods — that are adapted exclusively to chemosynthetic environments.

Finding multiple whale falls in proximity means the overlapping succession stages may allow species to migrate between carcasses, effectively turning the graveyard into a connected habitat rather than a series of isolated biological islands. Whether that dynamic is occurring at this newly discovered site remains a question researchers will need to address through further sampling.

Limits of what is currently known

The full taxonomic scope of what lives at the site is not yet established. Identifying potentially new species from submersible footage alone is preliminary; formal species descriptions require physical specimens, morphological analysis, and genetic sequencing.

The claims about novel organisms, while scientifically plausible given the inaccessibility of the deep sea, remain unconfirmed until peer-reviewed taxonomy is completed.

The precise location and depth of the graveyard have not been publicly disclosed in detail, which limits independent verification at this stage. Deep-sea expeditions of this kind are expensive and logistically complex, meaning follow-up dives to collect physical samples may take months or years to organize.

The research was first published in the journal Nature.

In a landmark achievement for robotics, a bicycle-riding robot has performed the world’s first unassisted front flip, astonishing engineers and robotics enthusiasts alike. The feat, accomplished by a team at the University of Tokyo’s JSK Robotics Laboratory, marks a significant leap in the capabilities of autonomous machines and could have far-reaching implications for robotics research, mobility, and future transportation.

Engineering a robotic stunt rider

The robot, dubbed “Rider-Bot,” is built to mimic the movements and balance strategies of human cyclists. Unlike previous robotic stunts that required ramps, wires, or external supports, Rider-Bot executed the front flip entirely on its own—using only its onboard sensors, actuators, and algorithms.

The successful flip took place on a flat surface, with the robot accelerating, launching, rotating 360 degrees forward, and landing upright—all without human intervention.

According to the research team, achieving this level of acrobatic control required a combination of advanced gyroscopes, real-time feedback loops, and machine learning techniques. “This is a major step toward robots that can adapt dynamically to complex environments and perform tasks that were previously considered impossible for machines,” said Dr. Yuta Sato, lead researcher on the project.

Implications for robotics and beyond

The successful front flip demonstrates not only the mechanical resilience of Rider-Bot but also the sophistication of its AI-driven decision-making. The robot’s control system anticipates and compensates for shifts in momentum and balance, much like a skilled human cyclist would.

Researchers believe this breakthrough could inform the design of next-generation delivery robots, autonomous vehicles, and even disaster-response machines capable of navigating unpredictable terrain. “Our work shows that robots can be taught to handle extreme maneuvers, which opens up many possibilities for robust, adaptable mobility in real-world scenarios,” Dr. Sato noted.

Challenges and next steps

Despite the successful demonstration, the team acknowledges ongoing challenges. Perfecting the landing and ensuring consistency across different environments remain top priorities. The researchers plan to enhance the robot’s learning algorithms and test it in outdoor and obstacle-rich settings.

The team also hopes their achievement will inspire further research into dynamic locomotion and the integration of robotics into everyday life. “We’re just beginning to scratch the surface of what intelligent, mobile robots can do,” said Dr. Sato.

A milestone for human-robot collaboration

The Rider-Bot’s unassisted front flip not only pushes the boundaries of robotics but also highlights the growing intersection between human athleticism and machine intelligence. As robots become more capable of complex movement, their potential roles in society—from urban mobility to emergency response—are poised to expand dramatically.

With this record-breaking feat, the University of Tokyo’s robotics team has set a new benchmark for autonomous machines, paving the way for future innovations in robotic agility and intelligence.

💾

France is advancing plans to develop its own military artificial intelligence (AI) infrastructure, codenamed Arcadia, aiming to reduce its reliance on American defense technology firms—chiefly Palantir Technologies.

Supported by the French Ministry of Armed Forces, the Arcadia initiative forms part of a broader strategy in Paris to safeguard operational autonomy over military data systems. The urgency for such sovereignty has heightened as NATO allies increasingly adopt US-developed AI platforms for critical command-and-control operations.

Why France is building its own military AI platform

The French government’s concerns are both strategic and structural. Palantir’s Gotham and Artificial Intelligence Platform (AIP), which consolidate battlefield data for predictive analysis and decision support, are now central to US military operations and are being actively promoted to European partners.

French defense officials have raised serious concerns about the risks associated with channeling classified operational data through foreign-owned systems, even those operated by close allies. Issues around data sovereignty—specifically, control over access, auditing, and the potential for denial of service—have taken center stage in France’s defense procurement debates.

The French initiative appears to draw a sharper line by opting to develop indigenous capability rather than negotiate data-handling clauses with a foreign vendor.

Technical architecture: What a domestic platform requires

Developing a viable alternative to Palantir’s military solutions is a complex, long-term undertaking. Palantir’s Gotham platform, for instance, integrates structured and unstructured data from a variety of sources—signals intelligence, geospatial inputs, logistics, and human intelligence—into a unified system that allows analysts to query and identify patterns across multiple domains simultaneously.

To replicate this, France needs more than a data fusion engine; the project demands sophisticated ontology management, granular role-based access control, and modular pipelines capable of real-time data ingestion from diverse sensors.

France’s defense procurement agency, the Direction Générale de l’Armement (DGA), is familiar with commissioning sophisticated defense systems. However, developing military-grade AI at this scale requires continuous software iteration—something traditional defense contractors may not be fully equipped to handle.

The French defense technology sector, featuring giants like Thales and Airbus Defence alongside a burgeoning group of dual-use AI startups, is expected to play a central role. Whether the platform will leverage open-source tools or rely on proprietary frameworks remains to be determined.

Interoperability risks with NATO allies

A major technical hurdle is interoperability. If US, UK, and other NATO allies are using Palantir-based platforms during joint missions, a French-developed system built on distinct data ontologies and APIs could complicate integration.

Effective intelligence sharing, collaborative operational planning, and coordinated targeting all rely on seamless data exchange. A sovereign French platform that cannot interface smoothly with allied infrastructure risks introducing new vulnerabilities, even as it addresses concerns about foreign control.

France’s experience with the development of its national military AI strategy suggests the DGA is aware of this tension, but resolving it will require either building compatible APIs from the start or negotiating data-exchange standards with NATO’s C3 architecture bodies — a process that can take years.

Procurement timeline and industrial capacity

No firm delivery timeline has been confirmed publicly, and that ambiguity itself carries operational risk. Military AI platforms require continuous retraining on current threat data, red-team adversarial testing, and integration with live classified networks — all of which take time even after initial software delivery.

Rushing a system to fielding without adequate validation can create the kind of brittle AI behavior that defense analysts warn can lead to catastrophic misclassification in high-tempo scenarios.

The French effort, as reported by Military AI, represents a calculated bet that sovereignty over defense AI infrastructure is worth the cost and delay of domestic development.

Whether Arcadia can match the operational maturity of a system that Palantir has refined through years of real-world deployment in US and allied theaters remains the central unanswered question.

A process originally developed to decarbonize steel production may also function as a large-scale electrical energy storage system, according to research into molten oxide electrolysis (MOE) and iron-based flow battery chemistry.

The convergence of these two approaches has drawn attention from engineers looking for ways to store renewable electricity without relying on lithium or other scarce materials.

How molten oxide electrolysis works in iron production

Molten oxide electrolysis involves dissolving iron ore — primarily iron oxide, or Fe2O3 — in a molten oxide electrolyte at temperatures exceeding 1,500°C (2,732°F). An electrical current is passed through the melt, reducing the iron oxide at the cathode to produce liquid metallic iron, while oxygen is released at the anode.

The process produces no direct carbon dioxide emissions, unlike the conventional blast furnace route that relies on coke as a reductant. Boston Metal has been among the companies advancing this approach at pilot scale, and the company’s progress on emissions-free steel illustrates how capital-intensive the transition away from fossil-fuel-based metallurgy will be.

Iron-air and iron flow batteries as a parallel technology

Separately, iron-based flow batteries and iron-air batteries have been gaining traction as candidates for long-duration energy storage. Iron-air cells cycle between oxidation states of iron — essentially rusting and de-rusting the electrode material — to store and release electricity.

Form Energy has publicly discussed iron-air battery systems targeting 100-hour storage duration at costs intended to compete with natural gas peaker plants. Iron flow batteries, meanwhile, use dissolved iron salts in an aqueous electrolyte circulated through electrochemical cells, with the energy stored in the oxidation state difference between iron(II) and iron(III) ions.

The dual-use concept: steelmaking reactor as energy storage

The more technically ambitious idea explored in recent research is whether a single electrochemical system could perform both functions — reducing iron ore while storing energy during periods of excess renewable generation, and releasing that energy on demand. In this model, the electrolysis cell charges when electricity is cheap and abundant, reducing iron oxide to iron metal.

When electricity is needed, the process reverses: the iron is re-oxidized, generating current. The iron ore feedstock effectively becomes the storage medium.

This approach draws on the well-understood electrochemistry of the Fe2+/Fe3+ redox couple, but scaling it to industrial steelmaking temperatures and throughput rates introduces significant materials challenges.

Electrode and refractory materials must survive prolonged exposure to both highly oxidizing and highly reducing conditions at extreme temperatures, and the round-trip energy efficiency of a reversible high-temperature system remains an open engineering question. Thermal losses alone at 1,500°C (2,732°F) represent a substantial efficiency penalty compared to room-temperature flow batteries.

Grid integration and practical constraints

From a grid-storage perspective, coupling an industrial process to electricity dispatch introduces scheduling conflicts. A steel plant optimized for continuous production cannot freely pause or reverse its electrochemical reactors to follow an electricity price signal without affecting throughput and product quality.

Partial decoupling — using dedicated MOE cells purely for storage, separate from production cells — would recover flexibility but at increased capital cost and facility footprint.

Iron as a storage material does offer one practical advantage over lithium-ion systems: it is abundant, geographically well-distributed, and carries no significant supply chain concentration risk.

The electrochemical reactions are also well-characterized, which lowers the barrier for modeling system behavior over thousands of charge-discharge cycles.

The path from laboratory electrochemistry to a commercially viable dual-use plant requires resolving electrode degradation rates, thermal management at scale, and the economic trade-offs between storage revenue and steel output.

💾

Engineers have developed a robotic arm that draws on the sensory architecture of the octopus to solve one of soft robotics’ persistent problems: how a machine can reliably grasp irregularly shaped objects without visual feedback alone.

The system integrates distributed tactile sensors across a flexible, multi-segment limb, allowing it to register contact force, surface geometry, and slip events simultaneously.

Why the octopus offers a useful mechanical model

The octopus arm is not simply a flexible appendage. Each of its eight arms contains roughly two-thirds of the animal’s total neurons, enabling semi-autonomous local processing of sensory data.

This distributed neural architecture means the animal does not need to route every tactile signal to a central brain before responding. Engineers working on dexterous manipulation have long recognized this as a functionally efficient design worth replicating in hardware.

The research team constructed their arm from soft elastomeric segments embedded with arrays of pressure-sensitive elements. Rather than relying on a single point of force measurement, the design distributes sensing across the full contact surface, producing a spatial map of pressure that updates continuously during a grasp.

Tactile sensor architecture and signal processing

The tactile sensors in the system function as piezoresistive or capacitive transducers — materials that change their electrical resistance or capacitance in proportion to applied mechanical load.

When arranged in a dense grid across the arm’s inner surface, they generate pressure maps with sufficient spatial resolution to distinguish, for example, the edge of a cylindrical object from a flat face.

Critically, the arm’s control architecture processes this tactile data locally before passing summary signals upstream. This reduces communication latency and means the limb can initiate corrective motion — tightening around a slipping object, for instance — faster than a centralized controller could respond.

The approach mirrors the ganglionic processing found in cephalopod arms, where reflex-like responses originate at the limb level rather than at the brain.

For engineers building manipulation systems for unstructured environments such as surgical robotics, subsea inspection, or warehouse automation, this matters because object geometry is rarely known in advance and visual occlusion is common once contact is made.

Tactile feedback fills that perceptual gap. Related work on soft robotic grippers has shown that compliance alone is insufficient without closed-loop sensory feedback to govern contact forces.

Structural compliance and material tradeoffs

Building a soft arm that is both compliant enough to conform to object surfaces and stiff enough to transmit useful gripping force is a recurring material tradeoff in the field. Elastomeric bodies deform under load, which helps distribute contact stress but also makes proprioception — knowing the arm’s own shape — difficult without additional strain sensors.

The research team addressed this by embedding shape-sensing elements alongside the tactile arrays, giving the controller a real-time estimate of limb configuration.

This dual-sensing approach adds fabrication complexity and raises questions about long-term durability. Soft elastomeric materials are susceptible to fatigue, tearing, and delamination at sensor integration points, particularly under repeated flexion cycles.

Whether the sensor-elastomer interface can withstand the cycle counts required for industrial deployment remains an open question the team has not fully resolved.

Constraints on deployment and scaling

The current prototype demonstrates proof-of-concept grasping across a range of object shapes, but several engineering constraints limit near-term application.

The actuators driving the soft segments — typically pneumatic or tendon-based — introduce their own latency and require external pressure sources or motor assemblies that offset the compactness the soft body provides. Miniaturizing these systems without degrading force output is a known challenge in soft robotics research.

Power consumption for continuous tactile sensing across large sensor arrays also scales non-trivially. In a system with hundreds of individual sensing elements, maintaining real-time readout requires dedicated multiplexing circuitry, adding weight and heat load that a fully untethered platform must manage within tight energy budgets.

A team of international researchers has uncovered new evidence that the Chicxulub meteorite impact, long known for causing the extinction of the dinosaurs, also created an underground environment that supported life for millions of years longer than previously thought.

This groundbreaking study not only reframes the narrative of Earth’s most infamous mass extinction event but also offers fresh insights into the origins of life on our planet and potentially others.

Meteorite impact: Catastrophe and opportunity

66 million years ago, a 6-mile-wide asteroid slammed into the Yucatán Peninsula, forming the Chicxulub crater and triggering a mass extinction that eradicated around three-quarters of Earth’s species, including all non-avian dinosaurs. The catastrophic event left a crater nearly 200 kilometers in diameter and profoundly altered the planet’s surface and subsurface environments.

Beneath the devastation, the collision’s immense heat fractured rocks, mingled with seawater, and set off the formation of a vast hydrothermal system—an underground network of hot water and porous rocks. These conditions, the researchers say, were ideal for sustaining microbial life, creating countless microenvironments that shielded organisms from surface extremes.

Expedition 364: Drilling for answers

In 2016, scientists from the International Ocean Discovery Program and International Continental Scientific Drilling Program embarked on Expedition 364 to drill into the crater’s peak ring. There, they collected samples of potassium-rich feldspar, a mineral formed from circulating hot fluids after the impact. Dr. Annemarie Pickersgill of the Scottish Universities Environmental Research Centre (SUERC) used advanced argon-argon dating techniques to analyze the feldspar.

The results were astonishing: the hydrothermal system persisted for at least eight million years—about four times longer than previous estimates. The team’s computer models, updated with new geological and physical data, confirmed that a combination of high rock permeability, sustained heat from the impact, and natural geothermal conditions allowed the underground system to endure far longer than once believed.

Implications for astrobiology and Earth’s origins

The study has wide-reaching implications. Dr. Pickersgill noted, “Wherever on Earth you find flowing warm water, you find life… we were still surprised by the outcomes of our research.” The findings suggest that impact-generated hydrothermal systems could have incubated early life on Earth and may exist elsewhere in the solar system.

Dr. Evangelos Christou, co-author and former PhD student at the University of Glasgow, highlighted how advances in computational modeling made it possible to simulate these complex systems with unprecedented accuracy. This research could inform the search for extraterrestrial life, particularly on planets like Mars, which experienced many large impacts during periods when water was more abundant.

Next steps in the search for life

The team stresses that impact craters may be prime locations for future astrobiological exploration. The porous rocks and underground heat could have provided refuge for life, both on early Earth and on other planets. As Dr. Pickersgill observed, “These findings could help future missions to other planets determine which impact craters might have been most likely to sustain life.”

The research brought together experts from institutions including the University of Glasgow, Purdue University, University of Texas at Austin, and several others, collectively advancing our understanding of how cataclysmic impacts can also spark conditions ripe for the emergence of new life.

The study was first published in the journal Communications Earth & Environment.

A Columbia University-led team has developed a revolutionary microscopy technology that could dramatically advance 3D tissue imaging while slashing costs and complexity.

The work, led by Professor Raju Tomer and published in Nature Biotechnology, introduces a new optical framework and modular microscope add-ons that could make high-resolution imaging accessible to labs and clinics worldwide.

Breaking barriers in 3D tissue imaging

High-resolution, three-dimensional images of intact tissues are crucial for mapping neural circuits, diagnosing disease, and training artificial intelligence systems.

However, traditional imaging tools force researchers to choose between expensive, shallow-depth, oil-immersion lenses or cheaper, deep-penetrating air lenses that blur when used with chemicals that clear tissue for 3D viewing.

The HySIL innovation

The Tomer team’s innovation, dubbed HySIL (Hybrid Solid–Liquid Optics), pairs a simple curved solid lens with a custom-matched immersion liquid, allowing the two to function as a seamless optical system.

This enables even inexpensive air lenses to deliver sharp, high-resolution images through centimeter-scale tissues, regardless of the common sample preparation method—without hardware changes.

To showcase the technology, the researchers built SCOPE, a modular device that attaches to existing light-sheet microscopes, and Super-SCOPE, a proof-of-concept variant offering even higher resolution.

These technologies have also been integrated into a compact, projector-based light-sheet microscope (pLSM) developed by Tomer’s group in 2024, which is now commercially available as SLICE.

“Our design breaks the long-standing trade-off between performance and accessibility in microscopy,” Tomer explained. “By making the immersion liquid an active optical component, we achieve the performance of elite systems at a fraction of the cost—suitable for teaching labs or clinics in low-resource settings.”

In collaboration with academic partners, the team demonstrated pLSM-SCOPE’s capabilities on whole mouse, salamander, and cavefish brains (for neural circuit mapping), lab-grown human brain tissues (for disease and development studies), and intact human cancer biopsies (for 3D pathology). HySIL is versatile and can enhance other types of microscopes, including confocal and two-photon systems.

Jack Glaser, co-founder and CEO of MBF Bioscience—an industry partner—praised the technology’s practical impact: “HySIL offers the rare combination of lower cost and higher performance. Engineering this into a reliable system makes advanced imaging broadly available to working labs.”

For decades, tissue analysis has largely relied on thin, two-dimensional slices, but the new technology enables affordable, accessible 3D imaging—crucial for visualizing the full architecture of tissues.

Hanina Hibshoosh, Professor of Pathology and Cell Biology at Columbia University Irving Medical Center, emphasized: “3D imaging lets you see entire tissue structures, not just cross-sections. Affordable tools like pLSM-SCOPE will be essential as AI expands our ability to analyze tissue data.”

A transformative step for research, medicine and AI

The research was supported by the National Institutes of Health, and Columbia University has filed patent applications on the technologies. Tomer has served as a consultant for MBF Bioscience, with several co-authors also employed there. Other Columbia contributors include Professors Rene Hen, Christopher Makinson, and Maria Tosches, as well as a team of postdoctoral fellows, doctoral candidates, and research staff.

Tomer concluded, “Access to dense, tissue-scale image datasets will transform biology and medicine as both increasingly depend on detailed imaging.”

Engineers have studied how jumping spiders see and used this unusual biological model to create a working 3D imaging device called SpiderCam. Unlike traditional depth cameras, SpiderCam avoids the usual bulk and high energy use. Their research could lead to smaller, more efficient sensors for robotics, medical devices, and autonomous systems where saving space and power is important.

How spider vision differs from conventional optics

Most jumping spiders have a layered retina in their main eyes that can capture images at different focal depths at the same time. Each layer is set at a different depth, so the same light creates several blurred images at once.

The spider’s nervous system uses the differences in blur between layers to figure out depth. It does this without needing two cameras, time-of-flight pulses, or special light projectors.

Most 3D cameras use one of three methods: comparing images from two separate sensors, measuring the time it takes for light to bounce back (like LiDAR), or projecting patterns of light onto a scene.

Each of these methods requires additional hardware, such as more lenses, more light sources, or wider spacing between sensors, which makes it hard to make them small. The spider’s way, called depth from defocus, uses just one eye and doesn’t need any of that extra equipment.

Translating biology into a functional sensor

The research team built a multi-layered optical system that copies the spider’s retina. They used a special lens and aperture setup to split light in a way that creates both sharp and blurred images on a stack of sensors. Then, computer processing turns the differences in blur into a detailed depth map.

As a result, this sensor works passively. It does not need to send out infrared pulses or project patterns, which means it uses much less power.

The elimination of an active illumination source also means the device works in outdoor daylight without the signal-washout problems that plague many IR-based depth sensors.

Calibration is a challenge for depth-from-defocus systems. The way blur relates to depth is very sensitive to small differences in the lens and how the device is made, so each unit may need its own calibration. This could make mass production more difficult unless the assembly process is highly precise.

Depth accuracy and working range trade-offs

Depth-from-defocus systems have a inherent limit on accuracy, determined by the lens opening and the sensor’s resolution. Bigger lens openings help measure depth more accurately up close, but they also shorten the range over which the system works well.

Because of this, the spider-inspired design has to be adjusted for a specific distance range. This means it is less flexible than LiDAR, which can measure depth over much longer distances.

This trade-off is fine for close-up applications, such as surgical cameras, drones avoiding nearby obstacles, or robots handling objects. But it does not work well for self-driving cars that need to see far ahead at high speeds.

Fabrication and integration outlook

Stacking photodetector layers with precisely spaced optical elements at small form factors requires fabrication processes compatible with existing CMOS foundry workflows — something the research team appears to have considered in the sensor’s design.

Whether the layered stack can be produced at wafer scale without accumulating alignment errors that degrade depth accuracy remains a central engineering question before any commercial iteration is feasible.

More broadly, this research shows that biological visual systems evolved to work well with very limited energy and space. This makes them good models for engineers who need to design efficient hardware under similar limits.

The jumping spider’s retina has been shaped by those pressures across millions of generations, and the depth-sensing solution it arrived at — passive, monocular, multi-focal — is one that silicon engineers are only now beginning to replicate with sufficient fidelity to be useful.

A research team has fabricated a functional computer chip using molybdenum disulfide (MoS2), a two-dimensional semiconductor material, integrating 1,400 transistors onto a single chip — a milestone in the effort to build AI-capable hardware beyond the limits of conventional silicon.

The work addresses a specific bottleneck in modern computing: the energy cost of running machine learning workloads on silicon-based devices. As AI models grow in parameter count, the gap between what silicon can efficiently deliver and what inference tasks demand has widened considerably.

What MoS2 brings to chip design

MoS2 belongs to a family of transition metal dichalcogenides — layered materials that can be exfoliated or grown down to a single atomic layer, typically around 0.65 nanometers thick. Unlike silicon, which degrades in electrical performance below approximately 5 nanometers, MoS2 retains a direct bandgap at the monolayer scale.

That property allows transistors built from it to switch cleanly between on and off states even at extreme thinness, which is the fundamental requirement for digital logic.

The bandgap in monolayer MoS2 sits at approximately 1.8 electronvolts (eV), compared to silicon’s 1.12 eV. A wider bandgap generally translates to lower leakage current — the parasitic drain that wastes energy in densely packed transistor arrays. For AI inference tasks running millions of switching cycles per second, that reduction in leakage carries direct power savings.

Fitting 1,400 transistors onto a single MoS2 chip is a meaningful step beyond prior demonstrations, which have typically remained in the range of tens to a few hundred devices.

The researchers reportedly achieved this by refining gate dielectric deposition — the insulating layer that sits between the gate electrode and the semiconductor channel — to reduce interface defects that would otherwise scatter charge carriers and degrade switching speed.

The fabrication challenge

Scaling 2D semiconductor devices is not a straightforward extension of silicon process flows. MoS2 films grown by chemical vapor deposition (CVD) frequently contain grain boundaries — interfaces between crystalline domains — that act as scattering sites and increase resistance. Achieving uniform coverage across a full wafer while controlling grain size remains an open problem in the field.

Contact resistance at the metal-to-MoS2 junction also poses a persistent obstacle. In silicon transistors, highly doped source and drain regions reduce this resistance to negligible levels.

In MoS2, the absence of established doping protocols means engineers must rely on work-function-matched metal contacts, a solution that works in laboratory devices but has proven difficult to reproduce at wafer scale.

These fabrication constraints put the 1,400-transistor chip in a different context from commercial silicon, where leading-edge nodes pack billions of transistors per square centimeter. The MoS2 result is better understood as a proof of circuit-level functionality rather than a density record.

Where this fits in the post-silicon roadmap

The semiconductor industry has operated under a de facto assumption that silicon will reach physical scaling limits sometime in the 2030s, prompting sustained investment in alternative channel materials.

MoS2 and other 2D materials appear in the International Roadmap for Devices and Systems (IRDS) as candidate channel replacements for sub-1nm nodes, precisely because their atomic thinness sidesteps the quantum tunneling problem that degrades silicon at those dimensions.

Whether MoS2 can transition from research chips to manufacturable products depends less on the physics, which appears favorable, and more on whether CVD growth processes can hit the defect-density targets that foundry-level yield requires. The 1,400-transistor result suggests circuit complexity is no longer the binding constraint — materials uniformity and contact engineering are.

If those process questions are resolved, ultra-thin MoS2 chips could eventually find a first deployment not in high-performance data center accelerators but in edge AI devices, where the combination of low leakage current and physical thinness aligns with strict power and form-factor budgets.

This work was first published in the journal Nature Electronics.

China has unveiled the Z-21, a heavy attack helicopter that military analysts are comparing directly to the Boeing AH-64E Apache Guardian, the United States Army’s primary rotary-wing strike platform. The aircraft’s emergence marks a concrete step in China’s effort to field a domestically produced heavy attack helicopter capable of operating in high-intensity contested environments.

Airframe design and propulsion

The Z-21 follows a tandem-seat, twin-engine configuration consistent with modern dedicated attack helicopter doctrine. Its airframe geometry — including stub wings for weapon pylons and a narrow fuselage cross-section — mirrors the layout optimized for survivability and aerodynamic performance in nap-of-the-earth flight.

The helicopter is reported to be powered by turboshaft engines developed domestically, a departure from earlier Chinese rotorcraft programs that relied heavily on foreign powerplant technology. Precise engine designation and shaft horsepower output have not been officially confirmed by the People’s Liberation Army (PLA).

Weapons and sensor suite

The Z-21 is equipped with a chin-mounted cannon and provisions for air-to-ground missiles, anti-tank guided munitions, and unguided rocket pods across four hardpoints on its stub wings. Reports indicate integration of a millimeter-wave fire control radar — analogous to the Longbow radar mast system on the AH-64E — enabling over-the-horizon target acquisition and engagement in degraded visibility conditions.

A multi-spectral targeting system mounted in a nose turret is expected to provide infrared and day/TV imaging for precision engagement. These are capabilities the AH-64E has carried operationally for over a decade, meaning the Z-21 would be entering the field at a level of sensor maturity the Apache already surpasses in software and network integration.

How it stacks up against the AH-64E Apache

The AH-64E Apache operates with Link 16 datalink compatibility, manned-unmanned teaming (MUM-T) with reconnaissance drones such as the MQ-1C Gray Eagle, and a target handoff system that allows crews to pass fire missions across platforms in near-real time.

Whether the Z-21 has achieved comparable network interoperability within PLA joint-force architecture remains unclear from available information.

On gross weight and payload, the Z-21 appears to fall in a similar class to the Apache’s roughly 10,400 lb (4,717 kg) maximum takeoff weight, though official Chinese specifications have not been released. This matters operationally: payload capacity directly constrains the missile and fuel load carried on a given mission profile, affecting both range and lethality.

Survivability features

The AH-64E incorporates radar-absorbent materials, infrared suppression on exhaust outlets, and a ballistically tolerant rotor system designed to sustain hits from 23 mm projectiles. It is not yet established whether the Z-21 meets comparable protection standards.

Engine reliability and maintenance logistics for domestically produced Chinese turboshafts also carry uncertainties, given the historical difficulties PLA aviation programs have faced in achieving consistent turbine performance at scale.

For readers interested in how rotary-wing platforms fit into broader strike architectures, the US Army’s Future Attack Reconnaissance Aircraft program offers relevant context on next-generation attack helicopter requirements being pursued in parallel.

Strategic context for PLA rotary aviation

The Z-21’s development fits within a wider PLA modernization effort that has prioritized closing capability gaps in close air support and anti-armor warfare. Taiwan Strait contingency planning, as well as high-altitude operations along the Sino-Indian border where the Z-10 medium attack helicopter has already deployed, likely informed the Z-21’s design requirements.

A heavier, more capable platform extends the PLA’s ability to conduct sustained attack helicopter operations in mountainous or fortified terrain where ordnance load matters.

Until the Z-21 completes operational testing and enters confirmed PLA service, claims about its combat effectiveness relative to the AH-64E remain speculative. The gap between a prototype photograph and a battle-ready weapons system in serial production is one that Chinese aerospace programs have historically taken years to close.

💾

Autonomous taxi services, which have scaled rapidly across major metropolitan areas in the United States and China, are poised to break into mainstream European transit networks following a decisive regulatory shake-up.

The European Union has moved to eliminate fragmented regional transport laws by introducing a unified “testbed” approach for autonomous vehicle validation. The streamlined framework allows driverless fleet operators to bypass highly bureaucratic country-by-country approval processes.

Backed by European transport ministers, this policy shift removes long-standing regulatory friction and clears the way for extensive autonomous trials across major European capitals this season.

Navigating Europe’s patchwork regulations

For years, autonomous tech firms have faced severe deployment delays across Europe compared to their international peers. While Waymo operates thousands of driverless rides in the US and Baidu’s Apollo Go blankets dozens of Chinese cities, Europe has lagged behind due to highly fractured cross-border safety rules.

Industry critics, including France’s high planning commissioner, have frequently lambasted the bloc for falling into regulatory stagnation while foreign operators built massive real-world data advantages.

The newly adopted testbed protocol directly addresses this imbalance. By establishing centralized baselines for safety, liability, and hardware evaluation, an autonomous platform approved in one EU member state can deploy test fleets across neighboring EU member states with minimal administrative delay.

This single-approval template transforms the financial calculus for mobility operators, allowing developers to aggregate real-world urban telemetry from a diverse mix of European driving conditions—ranging from high-speed roundabouts to historical narrow streets—without re-litigating safety certifications at every national border.

Global tech collides on European asphalt

The regulatory loosening has immediately triggered an aggressive land grab among global ride-hailing networks, domestic automakers, and international AI developers. Several cities are at the forefront of this transformation, each playing a distinct role in the unfolding autonomous vehicle landscape.

Zagreb, Croatia serves as Europe’s primary operational testbed. In a collaborative venture, Croatian startup Verne—backed by hypercar manufacturer Rimac—has partnered with Uber and Chinese AI leader Pony.ai to run a pilot fleet of 10 fully driverless robotaxis on public roads.

London, United Kingdom is set to become a highly contested arena. Three major tech syndicates are initiating field trials this year. These include US sector leader Waymo, British autonomy pioneer Wayve (partnered with Uber), and Chinese tech giant Baidu’s commercial unit, Apollo Go.

Madrid and Munich are witnessing rapid continental expansions. Chinese developer WeRide has partnered with Uber to run dynamic fleet testing in the Spanish capital, while German roads will see Uber vehicles integrated with Momenta and Autobrains AI control suites.

The urban-rural disconnect

Despite the regulatory momentum, industrial analysts warn that market forces and political goals in Europe remain sharply misaligned. Management consulting firms like Boston Consulting Group predict that by 2035, the global operational volume of robotaxis will reach roughly three million vehicles.

However, while China is projected to host 850,000 units and the US up to 350,000, Europe’s stringent municipal laws and cautious investment profiles are expected to limit the continent to just 120,000 active autonomous commercial vehicles.

A primary sticking point centers on geographic targeting. Robotaxi developers are intensely focused on high-density, lucrative downtown districts where ride-hailing demand peaks and software routing is highly profitable.

Conversely, local European policymakers are pressuring tech firms to deploy their fleets in suburban and rural outer rings to fill critical gaps left by shrinking public transportation budgets.

Reconciling this economic mismatch—while transitioning vehicle designs through highly specialized validation testing—will dictate whether Europe’s newly fast-tracked autonomous networks achieve commercial viability or remain restricted to hyper-localized pilot loops.

A major technological shift is underway in the maritime industry as researchers and engineers successfully deploy systems that allow traditional seafarers to operate commercial vessels entirely from land.

The breakthrough, spearheaded by Norway’s SINTEF Ocean and Kongsberg Maritime under the collaborative SEAMLESS project, comes alongside a major regulatory milestone: the International Maritime Organization (IMO) has officially adopted its first voluntary International Code of Safety for Maritime Autonomous Surface Ships (MASS Code).

Developed largely through Norwegian research, this framework establishes global safety and environmental guidelines for safely integrating artificial intelligence (AI) and remote-control technology into international shipping lanes.

Moving the bridge to the shore

The maritime sector faces a severe workforce squeeze. Industry data from the BIMCO ICS Seafarer Workforce Report estimated a global deficit of roughly 89,000 ship officers by 2026. Transitioning crews from deep-sea voyages to shore-based control centers offers an immediate remedy to both labor shortages and operational overhead.

Moving operations landward strips away the need for bulky onboard living quarters, galley systems, and heavy bridge structures. This enables the engineering of significantly smaller, streamlined hull designs. Smaller vessels yield a dramatic drop in energy consumption and emissions, making all-electric transport economically viable over short-sea shipping routes and inland waterways.

Furthermore, moving the bridge to dry land completely reshapes maritime labor demographics. Shore-based work hours align naturally with family life, making recruitment far easier. It also opens the profession to individuals with physical limitations. A worker using a wheelchair cannot easily navigate a traditional vessel deck, but they can seamlessly pilot an identical “bridge on land.”

Intelligent fleet management

A primary technical bottleneck for uncrewed shipping has been ensuring that shore-based operators are not overwhelmed by incoming telemetry. Within the 26-partner SEAMLESS initiative, engineers focused heavily on dynamic information filtering.

During standard operation, the shore crew’s displays show only highly aggregated, critical system data. This streamlined layout allows a single operator to safely monitor multiple autonomous vessels at once. However, if a ship detects an unexpected hazard or a mechanical anomaly, the system flags the operator.

The control interface instantly expands to mirror a full-scale tactical overview, matching the telemetry arrays used by pioneering uncrewed surface vessels like the 24-meter Reach Remote 1—the world’s first uncrewed surface vessel built specifically for deep-sea seabed mapping.

If multiple vessels face concurrent operational issues, the system’s fleet architecture handles hand-offs automatically, reallocating control of the secondary vessels to neighboring remote crews to ensure human judgment is always available where it is needed most.

From pilots to production

Autonomous cargo shipping is rapidly shifting from limited technology demonstrations into standard commercial service. The all-electric cargo vessel Marit, engineered to cross the Oslofjord entirely without an onboard crew, currently operates with a two-person safety team while awaiting final regulatory clearance.

With the IMO’s new safety code taking effect this season, industry leaders view these collaborative frameworks as an absolute necessity for economic survival. Autonomy is no longer an abstract engineering milestone; it is the primary mechanism for maintaining a competitive, safe, and decarbonized maritime trade network in a highly constrained global labor market.

China has unveiled the Z-21, a heavy attack helicopter that military analysts are comparing directly to the Boeing AH-64E Apache Guardian, the United States Army’s primary rotary-wing strike platform. The aircraft’s emergence marks a concrete step in China’s effort to field a domestically produced heavy attack helicopter capable of operating in high-intensity contested environments.

Airframe design and propulsion

The Z-21 follows a tandem-seat, twin-engine configuration consistent with modern dedicated attack helicopter doctrine. Its airframe geometry — including stub wings for weapon pylons and a narrow fuselage cross-section — mirrors the layout optimized for survivability and aerodynamic performance in nap-of-the-earth flight.

The helicopter is reported to be powered by turboshaft engines developed domestically, a departure from earlier Chinese rotorcraft programs that relied heavily on foreign powerplant technology. Precise engine designation and shaft horsepower output have not been officially confirmed by the People’s Liberation Army (PLA).

Weapons and sensor suite

The Z-21 is equipped with a chin-mounted cannon and provisions for air-to-ground missiles, anti-tank guided munitions, and unguided rocket pods across four hardpoints on its stub wings. Reports indicate integration of a millimeter-wave fire control radar — analogous to the Longbow radar mast system on the AH-64E — enabling over-the-horizon target acquisition and engagement in degraded visibility conditions.

A multi-spectral targeting system mounted in a nose turret is expected to provide infrared and day/TV imaging for precision engagement. These are capabilities the AH-64E has carried operationally for over a decade, meaning the Z-21 would be entering the field at a level of sensor maturity the Apache already surpasses in software and network integration.

How it stacks up against the AH-64E Apache

The AH-64E Apache operates with Link 16 datalink compatibility, manned-unmanned teaming (MUM-T) with reconnaissance drones such as the MQ-1C Gray Eagle, and a target handoff system that allows crews to pass fire missions across platforms in near-real time.

Whether the Z-21 has achieved comparable network interoperability within PLA joint-force architecture remains unclear from available information.

On gross weight and payload, the Z-21 appears to fall in a similar class to the Apache’s roughly 10,400 lb (4,717 kg) maximum takeoff weight, though official Chinese specifications have not been released. This matters operationally: payload capacity directly constrains the missile and fuel load carried on a given mission profile, affecting both range and lethality.

Survivability features

The AH-64E incorporates radar-absorbent materials, infrared suppression on exhaust outlets, and a ballistically tolerant rotor system designed to sustain hits from 23 mm projectiles. It is not yet established whether the Z-21 meets comparable protection standards.

Engine reliability and maintenance logistics for domestically produced Chinese turboshafts also carry uncertainties, given the historical difficulties PLA aviation programs have faced in achieving consistent turbine performance at scale.

For readers interested in how rotary-wing platforms fit into broader strike architectures, the US Army’s Future Attack Reconnaissance Aircraft program offers relevant context on next-generation attack helicopter requirements being pursued in parallel.

Strategic context for PLA rotary aviation

The Z-21’s development fits within a wider PLA modernization effort that has prioritized closing capability gaps in close air support and anti-armor warfare. Taiwan Strait contingency planning, as well as high-altitude operations along the Sino-Indian border where the Z-10 medium attack helicopter has already deployed, likely informed the Z-21’s design requirements.

A heavier, more capable platform extends the PLA’s ability to conduct sustained attack helicopter operations in mountainous or fortified terrain where ordnance load matters.

Until the Z-21 completes operational testing and enters confirmed PLA service, claims about its combat effectiveness relative to the AH-64E remain speculative. The gap between a prototype photograph and a battle-ready weapons system in serial production is one that Chinese aerospace programs have historically taken years to close.

💾