A mobile irrigation robot developed by researchers at the University of California, Riverside is challenging one of agriculture’s most persistent assumptions: that crops in the same field require the same amount of water.

By mapping soil moisture at the level of individual trees, the system reveals significant variation even between neighboring plants, suggesting that conventional irrigation methods may be systematically inefficient.

The findings point to a broader shift in agricultural robotics, where mobile sensing systems are replacing static infrastructure to deliver more granular, data-driven decisions.

From Field Averages to Tree-Level Precision

Traditional irrigation relies on fixed sensors and uniform watering schedules, operating on the assumption that conditions are relatively consistent across a field. The robot developed at UCR takes a different approach, scanning soil conditions continuously as it moves through orchards.

In field trials across citrus groves in California, the system detected sharp differences in water availability between adjacent trees, despite identical irrigation inputs. These variations were linked to differences in soil composition, where finer soils retained water more effectively than sandier patches.

The robot measures electrical conductivity in the soil – a proxy for moisture – and combines those readings with calibration data from a limited number of ground sensors. The result is a detailed moisture map that identifies both under-watered and over-watered areas.

This level of resolution allows irrigation to be adjusted at a much finer scale, turning what has traditionally been a field-wide estimate into a localized decision.

Reducing Waste and Managing Risk

The implications extend beyond water conservation. Overwatering can damage crops by depriving roots of oxygen and increasing susceptibility to disease, while also washing fertilizers deeper into the soil, where they can no longer be absorbed.

By identifying these imbalances, the system enables growers to maintain soil moisture within a narrower, optimal range. In testing, the model achieved high accuracy with relatively few calibration points, suggesting that widespread deployment may not require dense sensor networks.

This efficiency is significant in an industry where the cost of installing and maintaining sensors can limit adoption of precision agriculture technologies.

The approach also aligns with broader pressures facing agriculture, particularly in water-constrained regions. As drought conditions intensify, growers are increasingly forced to either reduce production or find ways to use water more efficiently.

Robotics Expands Beyond Automation

Unlike many agricultural robots focused on harvesting or crop monitoring, this system highlights a different role for robotics: acting as a mobile data layer that enhances decision-making rather than directly performing physical tasks.

The platform used in the study is capable of autonomous navigation, although it was manually operated during trials. Future versions are expected to operate independently, covering larger areas and integrating more closely with irrigation systems.

Several challenges remain before commercial deployment, including adapting the system to different crops, soil types, and environmental conditions. The relationship between surface measurements and deeper soil moisture also requires further refinement.

The development reflects a broader trend in robotics toward combining mobility with sensing and AI-driven analysis. By moving through environments rather than relying on fixed points, robots can capture variability that static systems miss.

In agriculture, where small differences in soil conditions can have large impacts on yield and resource use, that shift may prove particularly consequential.

If validated at scale, tree-level irrigation mapping could redefine how farms manage water – not as a uniform input, but as a variable resource tailored to each plant.

Skild AI has acquired the robotics automation business of Zebra Technologies, a move that signals a shift toward unified control systems for warehouse robotics rather than isolated deployments.

The deal includes Zebra’s Symmetry Fulfillment platform, a system designed to coordinate fleets of robots and human workers in logistics environments. By combining this orchestration layer with Skild AI’s general-purpose robotics model, the company is aiming to address one of the most persistent challenges in automation: fragmentation across hardware, software, and tasks.

The acquisition positions Skild AI to move beyond model development into full-stack deployment, where AI systems not only control individual robots but manage entire warehouse operations.

From Task Specific Automation to Generalized Control

Warehouse robotics has traditionally been built around specialized systems, with different robots programmed for picking, transport, or inspection. These systems often operate independently, requiring significant integration effort and limiting flexibility.

Skild AI’s approach centers on what it calls an “omnibodied” model, designed to operate across different robot types without being tailored to a specific form factor. In principle, this allows the same AI system to control humanoid robots, mobile platforms, and robotic arms without retraining for each configuration.

The addition of Zebra’s orchestration software extends this capability from individual robots to coordinated fleets. The Symmetry platform enables real-time task allocation, workflow management, and human-robot interaction, providing the infrastructure needed to deploy heterogeneous systems in live environments.

Together, the two technologies suggest a shift from programming robots individually to managing automation as a unified system.

Orchestrating Mixed Fleets at Scale

The combined platform is intended to support a wide range of robotic systems within a single warehouse. This includes autonomous mobile robots for material transport, robotic arms for packing, and potentially humanoid systems for more complex manipulation tasks.

Such an approach reflects the operational reality of modern logistics, where no single robot type can handle all tasks efficiently. Instead, performance depends on coordination between different systems and their integration with human workers.

By embedding AI at the orchestration level, Skild AI is attempting to create a layer that can dynamically assign tasks, optimize workflows, and adapt to changing conditions without requiring extensive reprogramming.

This model also creates a feedback loop: data collected from deployments can be used to improve the underlying AI system, potentially increasing performance across all environments where it is deployed.

A Push Toward End to End Automation

The acquisition highlights a broader industry trend toward end-to-end automation platforms. Rather than selling individual robots or software components, companies are increasingly positioning themselves as providers of complete operational systems.

This shift is driven in part by the limitations of current approaches. Many warehouses still require significant manual configuration to integrate different automation tools, and retrofitting facilities to accommodate specific robots can be costly and disruptive.

Skild AI’s strategy suggests an alternative path, where existing warehouses are adapted through software and orchestration rather than physical redesign. By combining a general-purpose AI model with a proven coordination platform, the company aims to reduce the complexity of deploying automation at scale.

The approach also aligns with efforts by companies such as Nvidia to build infrastructure for physical AI, where simulation, data, and control systems are integrated into cohesive platforms.

The success of this strategy will depend on whether a single AI layer can reliably manage diverse robotic systems in complex, real-world environments. While the concept of “any robot, any task” remains ambitious, the integration of orchestration and intelligence represents a step toward more flexible and scalable automation.

As logistics operators seek to increase efficiency without overhauling existing infrastructure, the ability to coordinate mixed fleets of robots may become a defining feature of next-generation warehouse systems.

Boston Dynamics has integrated a new generation of AI models from Google into its industrial inspection platform, marking a step toward more autonomous and context-aware robotics in real-world environments.

The update brings Google’s Gemini and Gemini Robotics-ER 1.6 models into Boston Dynamics’ Orbit AIVI-Learning system, which powers inspection workflows for robots such as Spot. The integration reflects a broader shift in robotics toward combining physical systems with advanced reasoning models capable of interpreting complex environments and making decisions in real time.

The rollout is already live for existing AIVI-Learning customers, with the company positioning the upgrade as a foundational improvement in how robots understand and monitor industrial sites.

From Detection to Interpretation

Industrial inspection has traditionally relied on rule-based systems that identify predefined objects or anomalies. The integration of Gemini introduces a different approach, where robots can analyze scenes more holistically and reason about what they observe.

Using the updated system, Spot can perform tasks such as reading gauges, assessing fluid levels, counting materials, and identifying safety hazards like spills or debris. These capabilities extend beyond simple detection, requiring the robot to interpret visual signals and determine their operational significance.

This shift is particularly important in environments where conditions are dynamic and difficult to model in advance. Rather than relying on static rules, the system can adapt to new scenarios, enabling broader deployment across facilities with varying layouts and equipment.

The addition of “transparent reasoning” features also allows operators to review how the system arrives at its conclusions, offering greater visibility into AI-driven decisions – a requirement that is becoming increasingly important in industrial settings.

Continuous Learning in Live Environments

A defining feature of the updated platform is its ability to improve over time through continuous data collection and model updates. The system operates as a cloud-connected service, allowing performance improvements to be deployed without interrupting operations.

This “zero-downtime” update model reflects a shift toward treating robotics systems as evolving software platforms rather than static hardware installations. As new data is collected from deployed robots, the models can be refined to better understand specific environments and use cases.

The approach, however, also introduces new considerations around data sharing. Customers using AIVI-Learning are required to share operational data with Boston Dynamics to enable ongoing model training, highlighting the growing role of data as a core component of robotics performance.

Toward Site Wide Intelligence

Boston Dynamics frames the integration as a move toward “site-wide intelligence”, where robots contribute to a unified understanding of industrial operations. By combining visual inspection data with higher-level reasoning, the system aims to provide insights across safety, maintenance, and logistics.

This aligns with a broader industry trend toward physical AI systems that integrate perception, reasoning, and action. Companies such as Nvidia have emphasized similar approaches, focusing on the convergence of simulation, AI models, and robotics hardware.

In practical terms, the upgraded system enables Spot to handle more complex inspection workflows, from monitoring equipment health to tracking material movement. The ability to interpret gauges and other analog instruments is particularly relevant in industries where digital integration remains incomplete.

The integration of Gemini into Boston Dynamics’ inspection platform highlights how quickly robotics is evolving from task-specific automation to more generalized, intelligent systems. By embedding reasoning capabilities directly into deployed robots, companies are beginning to close the gap between perception and decision-making.

The remaining challenge lies in scaling these systems across diverse environments while maintaining reliability and trust. As robots take on more responsibility in industrial settings, their ability to explain and justify decisions may become as important as their technical performance.

Google has introduced a new AI model designed to improve how robots understand and operate in real-world environments, targeting one of the most persistent limitations in robotics: the ability to reason beyond predefined instructions.

The model, Gemini Robotics-ER 1.6, focuses on what researchers describe as embodied reasoning – the capacity for machines to interpret visual inputs, plan sequences of actions, and determine when a task has been successfully completed. The update reflects a broader shift in robotics from systems that execute commands to those that can make context-aware decisions in dynamic settings.

The model is being made available to developers through Google’s AI tooling ecosystem, positioning it as part of a growing effort to standardize software layers for physical AI.

Moving from Perception to Reasoning

Robotics systems have historically relied on separate modules for perception, planning, and control, often requiring extensive engineering to connect them. Gemini Robotics-ER 1.6 attempts to unify these functions, allowing robots to process visual information and translate it directly into action.

The model improves spatial reasoning, enabling robots to identify objects, understand their relationships, and break tasks into smaller steps. It can also track objects across multiple viewpoints, combining inputs from different cameras to build a more complete understanding of an environment.

This multi-view capability is particularly relevant in real-world settings, where occlusion, clutter, and changing conditions can limit the effectiveness of single-camera systems. By integrating multiple perspectives, robots can maintain situational awareness even when parts of a scene are temporarily hidden.

Another key advancement is success detection. The model allows robots to evaluate whether a task has been completed correctly, reducing reliance on external validation or rigid programming. This is a critical requirement for autonomous operation, particularly in environments where tasks may need to be repeated or adjusted in real time.

Interpreting the Physical World

One of the more practical capabilities introduced in the model is the ability to read instruments such as gauges, meters, and digital displays. This function is particularly relevant for industrial and inspection applications, where robots must interpret physical indicators rather than purely digital data.

In collaboration with Boston Dynamics, the system has been applied to robots like Spot, which are used for facility monitoring. The model can analyze visual inputs, identify key components such as needles or numerical readouts, and calculate values with a high degree of accuracy.

Reported improvements in instrument reading performance suggest a significant step forward. Accuracy has increased from earlier levels of around 23% to over 90% in some scenarios, indicating that robots are becoming more capable of handling tasks that require precise interpretation of real-world signals.

The model also incorporates safety-aware reasoning, allowing robots to identify potential hazards and avoid unsafe interactions. This reflects an increasing emphasis on aligning robotic behavior with physical constraints, particularly as systems move into environments shared with humans.

Building a Software Layer for Physical AI

The release of Gemini Robotics-ER 1.6 highlights a broader trend toward treating robotics as a software problem as much as a hardware one. As companies race to develop humanoid and autonomous systems, the ability to generalize across tasks and environments is becoming a key differentiator.

Efforts by companies such as Nvidia and others have focused on simulation and training infrastructure, while Google’s approach emphasizes reasoning and decision-making at runtime. Together, these developments point toward a layered architecture for physical AI, where perception, reasoning, and control are increasingly integrated.

The remaining challenge is translating these capabilities into reliable real-world performance at scale. While models like Gemini Robotics-ER 1.6 demonstrate significant progress in controlled evaluations, deployment in complex environments will require further advances in robustness, data integration, and system design.

Google’s latest model suggests that robotics is entering a phase where intelligence is defined less by isolated capabilities and more by the ability to connect perception, reasoning, and action. As embodied AI systems become more capable of interpreting and responding to the physical world, the boundary between digital intelligence and physical execution continues to narrow.

The extent to which this translates into widespread adoption will depend on how quickly these systems can move from experimental demonstrations to dependable tools in industry and beyond.

Chinese robotics firm Unitree Robotics is preparing to launch its most affordable humanoid robot globally, a move that could test whether the category is beginning to transition from industrial experimentation to early consumer markets.

The company plans to debut its R1 humanoid robot through AliExpress, targeting customers in North America, Europe, Japan, and Singapore. With a starting price of around $4,000 in China, the R1 is among the lowest-cost humanoid robots introduced to date, positioning it closer to consumer electronics than traditional industrial machinery.

The rollout comes as Unitree accelerates production and expands internationally, following a year in which it shipped more than 5,500 humanoid robots – far exceeding most global competitors.

Lower Prices Meet Global Distribution

The R1 reflects a broader push to reduce the cost of humanoid robotics while expanding access through global distribution platforms. By launching on AliExpress, Unitree is bypassing traditional enterprise sales channels and testing direct-to-market demand.

The robot stands just over 1.2 meters tall and is designed for dynamic movement, including running, recovering from falls, and performing coordinated motions. Marketed as “sport-ready”, it highlights Unitree’s focus on mobility and mechanical performance rather than immediate utility in structured work environments.

The pricing strategy marks a significant departure from earlier humanoid systems, which have typically been priced in the tens of thousands of dollars or higher. Even companies such as Tesla have suggested that future humanoid robots could cost around $20,000, placing Unitree’s offering well below that threshold.

The question is not only whether such pricing is sustainable, but whether it will translate into meaningful adoption beyond research labs and demonstration use cases.

Scaling Production Ahead of Demand

Unitree’s global expansion is closely tied to its manufacturing scale. The company has set a target of shipping between 10,000 and 20,000 robots in 2026, building on its current position as one of the highest-volume producers of humanoid systems.

According to industry estimates, competitors such as Figure AI and Agility Robotics have shipped only a few hundred units each, underscoring the gap between Chinese and U.S. production capacity.

Market research firm TrendForce expects Unitree to account for a substantial share of global humanoid output in the near term, reflecting both aggressive scaling and a focus on cost reduction.

At the same time, the company is preparing for a potential IPO in Shanghai, aiming to raise capital to expand manufacturing and research. The R1’s international debut may therefore serve a dual purpose: generating revenue while demonstrating global demand to investors.

From Demonstration to Early Adoption

The launch also highlights a shift in how humanoid robots are being positioned. Rather than targeting a single industrial application, the R1 appears designed as a general-purpose platform that can showcase capabilities and attract a broader user base.

Unitree has previously gained visibility through high-profile demonstrations, including coordinated performances by its robots on national television. The move into global e-commerce suggests a transition from spectacle to early commercialization, even if practical use cases remain limited.

For now, most humanoid robots are still used in research, education, and controlled environments. The introduction of a lower-cost model does not immediately resolve challenges around autonomy, reliability, or real-world utility.

However, it may begin to reshape expectations. If consumers and small businesses can access humanoid robots at a fraction of previous costs, the market could shift from a handful of experimental deployments to a larger base of exploratory use.

Unitree’s R1 launch represents one of the clearest attempts to test that transition. By combining lower pricing with global distribution, the company is effectively probing whether humanoid robotics can move beyond early adopters and into a broader commercial category.

The outcome will depend less on technical capability alone and more on whether users find meaningful ways to integrate these systems into everyday environments. For an industry still searching for its first large-scale application, that question remains open.

The development of humanoid robots is increasingly dependent not just on hardware breakthroughs or AI models, but on a growing global workforce capturing the physical world on camera. Across more than 50 countries, gig workers are now filming themselves performing everyday household tasks to generate training data for robots that are still years away from widespread deployment.

The model, led by startups such as Micro1, reflects a broader shift in how physical AI systems are built. Just as large language models relied on vast corpora of text scraped and labeled at scale, humanoid robots require detailed recordings of human interaction with objects in real-world environments. The difference is that this data must be created, not collected – and it is being produced inside people’s homes.

Building the Data Layer for Physical AI

Humanoid robots face a fundamentally different challenge from software-based AI systems: they must operate in unstructured, unpredictable environments. Tasks such as folding laundry, loading dishwashers, or organizing shelves involve subtle variations that are difficult to simulate or script.

To address this, companies are assembling large datasets of human activity, capturing how people manipulate objects in real settings. Workers are paid to record themselves performing routine tasks, often wearing cameras that track hand movements, object interactions, and spatial context.

The resulting footage forms the foundation for training robot perception and control systems. Companies such as Scale AI have already accumulated tens of thousands of hours of such material, while platforms like DoorDash have begun experimenting with allowing gig workers to contribute training data alongside their primary work.

This emerging pipeline suggests that physical AI will depend on a new category of data infrastructure – one that extends beyond digital content into the physical behaviors of human workers.

A Familiar Economic Structure in a New Domain

The economics of this system closely resemble earlier phases of the AI industry. Workers contributing data are typically paid hourly rates that are competitive within their local economies but represent a small fraction of the value generated downstream.

Participants receive no ownership over the data they produce and no share in the long-term value of the models trained on it. As humanoid robotics companies attract billions in investment, the gap between capital allocation and labor compensation is becoming more pronounced.

This structure mirrors the development of computer vision and natural language processing systems, where data labeling and annotation were outsourced globally. The key difference is that physical AI requires more invasive forms of data collection, capturing not just digital inputs but lived environments.

The result is a new layer of the gig economy, one that sits beneath the visible robotics industry and provides the raw material for its progress.

Privacy Risks Move Into the Home

Unlike earlier data pipelines, which largely relied on public or platform-generated content, the data used to train humanoid robots is often recorded in private spaces. Videos include kitchen layouts, household items, and other details that collectively form a detailed map of domestic life.

This raises questions about data ownership, consent, and long-term storage. Workers may have limited visibility into how their recordings are used, whether they are anonymized, or how long they are retained. The implications extend beyond individual privacy to broader concerns about the creation of large-scale visual datasets of private environments.

Researchers in human-centered computing have emphasized the need for clearer disclosure and safeguards, but industry practices remain inconsistent. As the volume of collected data grows, so too does the potential risk associated with breaches, misuse, or secondary applications.

The reliance on gig workers to generate training data underscores a central reality of humanoid robotics: progress depends not only on engineering advances, but on access to large-scale, real-world human behavior.

This data-centric approach may accelerate development, but it also introduces new questions about labor, ownership, and privacy. As physical AI moves closer to commercial deployment, the systems being built will increasingly reflect not just technological innovation, but the global infrastructure of work that supports them.

Robots have made rapid progress in vision and motion, but touch has remained a persistent limitation. Without reliable tactile feedback, even advanced systems struggle to handle fragile objects or safely interact with humans. A new class of flexible sensors developed by researchers at Penn State suggests that gap may be narrowing.

The team has created a lightweight “robotic skin” capable of detecting extremely small pressure changes while maintaining durability under repeated use. The development reflects a broader push in robotics to move beyond perception and mobility toward physical intelligence – systems that can interpret and respond to the physical world with greater nuance.

Turning Pressure into Real Time Control

At the core of the system is a small, flexible sensor built around graphene aerogel, a porous material that converts mechanical pressure into electrical signals. The structure allows the sensor to respond quickly to light touch while remaining stable under heavier loads, addressing a common tradeoff between sensitivity and durability.

Each sensor can register contact in just over 100 milliseconds and recover shortly after, enabling near real-time feedback. When arranged in arrays, these sensors generate pressure maps that function similarly to human skin, allowing robots to interpret how force is distributed across their surface.

This capability shifts tactile sensing from passive measurement to active control. In demonstrations, robotic hands equipped with the sensors adjusted grip strength dynamically, preventing damage to delicate objects such as soft food items. The system effectively translates touch into immediate motor responses, closing a loop that has historically been difficult to achieve in robotics.

From Grasping to Perception

Beyond simple force control, the sensor system introduces a new layer of perception. By analyzing pressure patterns, robots can begin to distinguish between different materials and objects based on how they respond to touch.

In experimental tests, researchers trained a lightweight model to classify food items using tactile data alone. After repeated training cycles, the system achieved accuracy above 99%, suggesting that touch-based recognition could complement or, in some cases, substitute for visual input.

This has implications for environments where vision is unreliable, such as cluttered industrial settings or domestic spaces with variable lighting. It also aligns with a growing interest in multimodal AI systems that combine vision, language, and physical interaction.

The same sensing approach has also been applied to wearable devices, where it can track pulse signals and joint movement with consistent accuracy. This points to potential crossover applications in healthcare, prosthetics, and rehabilitation.

Expanding the Role of Tactile Intelligence

The development highlights a broader shift in robotics toward integrating sensing, control, and learning into unified systems. While vision-based AI has dominated recent advances, tactile intelligence is emerging as a critical component for real-world deployment.

Companies such as Tesla and Nvidia have emphasized the importance of physical interaction in next-generation AI systems, particularly in humanoid robotics and automation. However, progress in touch sensing has lagged behind advances in perception and planning.

The Penn State research suggests that scalable, low-cost tactile systems may begin to close that gap. The sensors can also detect pressure changes in non-robotic contexts, such as monitoring swelling in battery systems – an early indicator of potential failure in electric vehicles.

Despite the progress, the technology remains in an early stage. Challenges include miniaturization, long-term reliability, and integration with existing robotic platforms. Researchers are also exploring ways to expand the sensing capabilities to include temperature and stretch, bringing the system closer to the complexity of human skin.

The ability to sense and respond to gentle touch is likely to be a defining feature of next-generation robots, particularly as they move into homes, healthcare settings, and collaborative workplaces. While the current system is still experimental, it illustrates how advances in materials science and AI are converging to address one of robotics’ most persistent limitations.

If scaled successfully, tactile sensing could shift robots from rigid, pre-programmed machines to adaptive systems capable of interacting with the physical world in a more human-like way.

The idea of robots operating inside the human body has long been associated with science fiction. But recent advances in DNA-based nanotechnology are beginning to translate that vision into early-stage experimental systems, where programmable molecular machines can move, sense, and interact with biological environments.

Researchers are now designing DNA “robots” capable of delivering drugs directly to diseased cells and identifying viral threats within the bloodstream. While these systems remain far from clinical deployment, they represent a shift in how robotics is defined – extending from mechanical systems into the molecular domain.

Reimagining Robotics at the Molecular Scale

Unlike conventional robots built from metal, electronics, and actuators, DNA robots are constructed from strands of nucleic acids that can be folded, connected, and programmed into functional structures. Using techniques inspired by origami, scientists can create rigid joints, flexible linkages, and dynamic components that mimic mechanical systems at a nanoscale.

This approach adapts established principles from traditional robotics – including rigid-body motion and compliant mechanisms – into a biochemical context. The result is a new class of machines that operate not through motors or gears, but through chemical interactions and structural transformations.

Controlling these systems presents a fundamental challenge. At the molecular level, motion is dominated by random thermal fluctuations, known as Brownian motion, which can disrupt precise behavior. To address this, researchers rely on biochemical programming methods such as DNA strand displacement, where specific sequences act as triggers to initiate movement or change configuration.

External signals, including light, magnetic fields, and electric fields, can also be used to guide these nanorobots, providing an additional layer of control in otherwise unpredictable environments.

Medical Applications Remain Experimental

The most immediate interest in DNA robotics lies in medicine, where the ability to operate at cellular or even molecular resolution could enable highly targeted interventions. In experimental settings, DNA robots have been designed to locate specific cell types, release therapeutic payloads, and potentially capture or neutralize viruses.

Such systems could function as “nano-surgeons”, delivering drugs with far greater precision than conventional treatments and reducing side effects associated with systemic therapies. Researchers are also exploring their potential to detect and bind to viral particles, including pathogens similar to COVID-19, as a step toward autonomous diagnostic or therapeutic platforms.

Beyond medicine, DNA robots may also serve as tools for nanoscale manufacturing. By positioning molecules and nanoparticles with sub-nanometer precision, they could enable new forms of computing and materials engineering that are difficult to achieve with existing fabrication techniques.

However, most current systems remain proof-of-concept demonstrations. They typically operate in controlled laboratory conditions and lack the robustness required for real-world biological environments.

From Proof of Concept to Scalable Systems

The transition from experimental prototypes to practical applications presents several challenges. In addition to environmental unpredictability, researchers face limitations in modeling and design. There is currently no comprehensive database of DNA mechanical properties, and simulation tools for predicting nanorobot behavior remain underdeveloped.

Scaling these systems will likely require advances across multiple domains, including bio-manufacturing, materials science, and artificial intelligence. Proposed approaches include the development of standardized DNA component libraries and the use of AI-driven design tools to optimize structures and predict performance.

The broader implication is that robotics may increasingly extend beyond traditional hardware into programmable biological systems. DNA robots, if successfully scaled, could redefine automation at the smallest possible level – enabling machines that operate not in factories or warehouses, but within cells and molecules themselves.

For now, the technology remains in its formative stage. But its trajectory suggests that the next phase of robotics innovation may be less about building larger, more capable machines, and more about engineering systems that can function where conventional robots cannot reach.

LG CNS is deepening its push into physical AI through a set of partnerships with Silicon Valley robotics startups, signaling a shift from enterprise software toward integrated AI and robotics systems. The move reflects a broader trend among large technology firms seeking to secure both the software intelligence and hardware platforms required for real-world automation.

The South Korean company announced that it has partnered with U.S.-based startups Config and Dexmate following its Open Innovation Summit held in Silicon Valley on March 19. The initiative is part of an ongoing effort to identify early-stage technologies that can be incorporated into LG CNS’s enterprise-focused AI and automation offerings.

Combining Robot Foundation Models with Hardware

At the center of the partnerships is Config, a startup focused on robot foundation models, a category of AI systems designed to generalize across tasks in physical environments. The company’s technology enables robots to learn from human motion data, translating real-world demonstrations into structured training inputs for robotic systems.

LG CNS plans to integrate Config’s models into its robotics stack to improve precision in dual-arm manipulation, a capability widely seen as critical for industrial automation and service robotics. Unlike traditional robotic programming, which relies on predefined instructions, these models aim to allow robots to adapt to variable environments with less manual configuration.

The partnership with Dexmate, meanwhile, extends LG CNS’s reach into hardware. Dexmate develops humanoid robots equipped with dual arms and wheel-based mobility, offering an alternative to bipedal locomotion that can simplify stability and deployment in structured environments.

LG CNS had previously invested in Dexmate, and the expanded partnership suggests a longer-term strategy of aligning software capabilities with specific hardware platforms rather than remaining hardware-agnostic.

Expanding the Definition of Physical AI

The company’s approach reflects an evolving definition of physical AI, where progress depends on the interaction between machine learning models and mechanical systems rather than advances in either domain alone. By working with both a model developer and a hardware manufacturer, LG CNS is positioning itself within a growing ecosystem that spans perception, control, and actuation.

This mirrors broader industry developments led by companies such as Nvidia, which has promoted integrated frameworks combining simulation, AI training, and robotics deployment. The emphasis on full-stack systems is becoming increasingly important as robotics moves from controlled demonstrations to operational environments.

LG CNS’s expansion into wheel-based humanoid systems also suggests a pragmatic approach to deployment. While bipedal robots remain a long-term goal for many developers, hybrid designs that prioritize stability and efficiency are gaining traction in logistics, manufacturing, and service applications.

Open Innovation as a Scaling Strategy

The partnerships were announced as part of LG CNS’s broader open innovation program, which seeks to identify and collaborate with startups at an early stage. This model allows large enterprises to access emerging technologies without building all capabilities in-house, while giving startups a pathway to commercial deployment.

For LG CNS, the strategy appears aimed at accelerating its transition from enterprise IT services into a provider of AI-driven automation infrastructure. By combining internal capabilities with external innovation, the company is attempting to build a flexible ecosystem that can adapt as both AI models and robotics hardware continue to evolve.

The challenge, as with much of the physical AI sector, lies in translating technical capability into scalable, real-world use cases. While partnerships can accelerate development, widespread deployment will depend on whether these integrated systems can deliver consistent performance in complex environments.

The planned public listing of Unitree Robotics marks a turning point for the humanoid robotics sector, which has long been defined by prototypes, research funding, and speculative timelines. By moving toward an initial public offering, the Hangzhou-based company is positioning itself as one of the first large-scale tests of whether humanoid robots can sustain a viable commercial market.

Unitree filed to list on the Shanghai Stock Exchange on March 20, seeking to raise 4.2 billion yuan, or about $610 million, to expand manufacturing and research. The company’s trajectory, from viral demonstrations to profitability within a year, places it at the center of a broader shift in how robotics companies are financed and evaluated.

Profitability Arrives Ahead of Mass Adoption

Unlike many peers, Unitree enters the public markets with profitability already established. The company reported an adjusted net profit of 600 million yuan in 2025, a sharp increase from its first profitable year in 2024. Revenue rose to 1.71 billion yuan from 392 million yuan the previous year, reflecting both volume growth and expanding product adoption.

This distinguishes Unitree from earlier entrants such as UBTech Robotics, which has remained unprofitable despite going public. The contrast highlights a widening gap between companies still operating in development mode and those beginning to scale production.

Even so, the market remains early. More than 100 humanoid robotics companies currently operate in China, according to Counterpoint Research, with consolidation expected as capital markets begin to impose stricter performance expectations. Unitree’s IPO is likely to serve as an early signal of which business models can sustain investor confidence.

From Quadrupeds to Humanoids

Unitree’s growth has been driven in part by a transition from quadruped robots to humanoid systems. The company shipped more than 30,000 quadrupeds between 2022 and 2025, establishing a hardware and supply chain base before scaling humanoid production.

In 2025, it sold 5,500 humanoid robots, which accounted for over half of its core revenue, up from less than 2% two years earlier. The majority of these units were sold to research institutions and educational users, indicating that widespread enterprise deployment remains limited.

The shift reflects a broader industry pattern, in which quadruped platforms have served as an intermediate step toward more complex humanoid systems. These earlier products provide revenue, operational data, and manufacturing experience that can be transferred into humanoid development.

Falling Prices and Vertical Integration

One of the more notable signals in Unitree’s prospectus is the rapid decline in pricing. The average price of its humanoid robots fell from roughly 593,400 yuan in 2023 to 167,600 yuan in 2025, bringing systems closer to a range that could support broader adoption.

At the same time, gross margins improved to nearly 60%, suggesting that cost reductions are being driven by manufacturing efficiencies rather than discounting alone. Unitree attributes this to its strategy of developing and producing key components in-house, reducing reliance on external suppliers.

This combination of falling prices and improving margins remains rare in the humanoid robotics sector, where most companies are still managing high costs and limited production volumes.

However, external dependencies remain. Like many robotics developers, Unitree relies on computing platforms and chips from Nvidia for core processing capabilities, leaving part of its supply chain exposed to geopolitical and trade uncertainties.

A Market Signal for Physical AI

Unitree’s IPO arrives amid intensifying global competition in humanoid robotics. In the United States, Elon Musk has said that Tesla plans to begin retail sales of its Optimus robots by 2027, framing humanoids as a future mass-market product.

At the same time, the concept of “physical AI” – systems that combine machine learning with real-world interaction – is gaining traction across the industry. Unitree’s robots were featured alongside other platforms at a recent conference led by Jensen Huang, underscoring growing alignment between hardware manufacturers and AI infrastructure providers.

Despite this momentum, near-term demand remains concentrated in research, education, and controlled industrial environments. Unitree’s own projections, which include plans to produce tens of thousands of humanoids annually within five years, suggest confidence in scaling, but not necessarily immediate mass adoption.

The company’s public listing will therefore function as more than a financing event. It will offer one of the first measurable indicators of whether investors view humanoid robotics as an emerging industrial category or as a longer-term technological bet.