iREX 2025: The Quiet Shift from Programmed to Perceptive
For decades, the international robotics industry has operated on a simple, reliable premise: a robot does exactly what it is told, in exactly the order it is told, and it does so with relentless consistency. The annual iREX exhibition in Japan has historically been the stage where that premise is celebrated, refined, and commercialized. But the 2025 edition of the show, held in December, signaled something different. The conversation has moved from programming to perception. The industry’s most established players are no longer just selling precision; they are selling the ability to understand, react, and adapt in real time.
This shift is not a rumor or a distant research aspiration. It is a commercial reality, evidenced by concrete announcements from the world’s largest robot manufacturer and a visible influx of Chinese competitors offering capabilities that were previously the domain of academic labs. For buyers, system integrators, and manufacturing executives, the implications are substantial. The hardware is becoming a platform, and the software is becoming the differentiator.
The announcement
The most significant news to emerge from the iREX 2025 floor was not a new robotic arm with a faster cycle time. It was a strategic partnership announced by Japan’s largest robot manufacturer—a company historically known for its closed, proprietary robot controllers—with NVIDIA. The timing was deliberate, coming just before the iREX exhibition opened its doors. The goal of this alliance is to accelerate what the industry now calls "physical AI" in industrial robotics.
Physical AI is a term that describes the fusion of artificial intelligence judgment with robotic hardware. It moves beyond simple pre-programmed sequences and into a realm where the machine can recognize its environment, make decisions based on that recognition, and execute those decisions in real time as the surroundings change. The partnership is built on an open-platform approach. The manufacturer explicitly stated that the intention is to "enhance connectivity with external AI environments" and to speed up the implementation of physical AI. In practical terms, this means connecting robot arms to modern AI software ecosystems that were previously incompatible with the factory floor.
This is a notable departure from the company’s historical strategy. For years, the value proposition of Japanese industrial robots was their reliability, which was achieved in part through tightly controlled, closed-loop systems. The controller was the brain, and it was proprietary. Opening that architecture to external AI environments is a strategic acknowledgment that the future of automation lies not in isolated islands of precision, but in interconnected systems that can learn and adapt.
The announcement was not just a memorandum of understanding. It included a concrete technical integration. The manufacturer has integrated NVIDIA’s Isaac Sim simulation platform into its workflow. This is a physics-based simulation environment used for testing and training AI-driven robots. Crucially, the manufacturer has contributed official 3D models of its robots to the Isaac Sim platform. This is a significant step for system integrators and AI developers, as it provides a validated, accurate digital twin of the physical hardware. These models can be used for virtual testing of new applications and for training AI models in a simulated environment before they are ever deployed on a physical factory floor.
The message from the manufacturer is clear: the era of the dumb, blind arm is over. The new era is one where the arm is a peripheral device for a larger, intelligent system. By providing official 3D models and integrating with a leading simulation platform, they are lowering the barrier to entry for AI development in industrial settings. They are effectively saying that their hardware is ready to be the hands for the AI brains being developed by the broader software community.
Product and availability details
While the NVIDIA partnership was the headline, the broader iREX 2025 floor told a more complex story about the state of the market. The most visible Chinese exhibitors at the show were not competing head-to-head with Japanese incumbents on traditional six-axis articulated robots. Instead, they concentrated their presence in market segments where Japanese players are either less visible or less aggressive on price. These segments were identified as low-cost collaborative robots (cobots), humanoid robots, and advanced tactile sensing.
This strategic positioning is telling. The Chinese manufacturers are not trying to out-engineer the Japanese on raw precision; they are trying to out-innovate them on accessibility and new form factors. Low-cost cobots, for instance, target small and medium-sized enterprises that may have been priced out of automation previously. Humanoid robots, while not yet ubiquitous in factories, represent a long-term bet on general-purpose automation that can navigate human-centric environments. Advanced tactile sensing is a critical enabler for tasks that require a delicate touch, such as assembly of small components or handling of deformable materials—tasks that traditional force-torque sensors struggle with.
Among these Chinese exhibitors, AgiBot stood out. Recognized as one of China’s leading humanoid robot makers, AgiBot used the iREX platform to make a significant market announcement: its official entry into the Japanese market. This is a bold move, as Japan is the home turf of the world’s most established robotics manufacturers. Entering this market requires not only competitive hardware but also a robust understanding of local safety standards, integration practices, and customer service expectations.
To support its entry, AgiBot showcased its vision-language-action (VLA) model, named ViLLA. A VLA model is a type of AI that takes visual input (what the robot sees), combines it with language instructions (what the user wants), and generates actions (what the robot does). This is a departure from traditional robot programming, which requires a programmer to explicitly define every movement. With a VLA model, the robot can interpret a high-level command like "pick up the red object and place it in the tray" by understanding the visual scene and executing the task without step-by-step coding.
The availability of these technologies varies. The NVIDIA Isaac Sim integration and the official 3D models are available now for developers and integrators to use in virtual environments. The specific commercial terms of the strategic partnership between the Japanese manufacturer and NVIDIA were not fully disclosed in the source material. Similarly, while AgiBot announced its entry into the Japanese market, the specific models, pricing, and launch dates for its humanoid robots in Japan were not detailed. What is known is that the VLA model ViLLA was demonstrated at the show, indicating a level of technical maturity that moves beyond concept.
It is also important to note what is not disclosed. The source material does not specify the exact timeline for the full rollout of physical AI capabilities on the Japanese manufacturer’s standard product line. It does not provide specific performance metrics for the VLA model, nor does it detail the pricing structure for the low-cost cobots from Chinese exhibitors. Buyers should expect a period of transition, where simulation tools and AI integration are available, but the full ecosystem of certified applications and support services is still maturing.
What it means for buyers
For a manufacturing executive or a system integrator, the developments at iREX 2025 are a signal to reassess long-term automation strategy. The shift from programmed to perceptive robots is not a gimmick; it is a fundamental change in how automation value is delivered.
First, the open-platform approach from the dominant Japanese manufacturer is a direct answer to a long-standing pain point: the difficulty of integrating modern AI with industrial hardware. In the past, if a company wanted to use a neural network to optimize a pick-and-place operation, they often had to build custom middleware to bridge the gap between the AI software and the proprietary robot controller. This was expensive, time-consuming, and brittle. The new partnership with NVIDIA, and the integration of Isaac Sim, changes this calculus. By providing official 3D models and opening the controller to external AI environments, the manufacturer is effectively commoditizing the hardware layer. The value is now in the software and the AI model that runs on top of the hardware.
For buyers, this means that the barrier to implementing AI-driven automation is lower than it has ever been. Virtual testing in Isaac Sim allows for rapid prototyping of new cells without risking physical assets. AI models can be trained in simulation, where failures are free, and then transferred to the physical robot. This reduces the risk and cost associated with deploying new automation solutions. It also means that the pool of potential solution providers is broader, as smaller AI startups can now develop applications for a major industrial arm without needing deep expertise in proprietary controller languages.
Second, the entry of Chinese players like AgiBot into the Japanese market—and the broader visibility of low-cost cobots and humanoid robots—is a price and innovation pressure valve. For buyers, this is beneficial. It creates a wider range of options at different price points. The presence of low-cost cobots challenges the incumbents to justify their premium pricing with superior service, reliability, or performance. The focus on humanoid robots and tactile sensing, while still nascent, points to a future where robots can handle a wider variety of tasks, including those that require human-like dexterity and perception.
However, buyers must also exercise caution. The source material does not provide specific data on the reliability, support infrastructure, or long-term viability of the new entrants. The announcement of a VLA model like ViLLA is impressive, but the practical deployment of such models in a factory setting requires robust data pipelines, edge computing infrastructure, and fail-safe mechanisms. The source material does not specify how AgiBot will handle service and support in Japan, nor does it provide details on the safety certifications of the low-cost cobots.
The strategic partnership with NVIDIA is a strong indicator of direction, but it does not guarantee immediate, plug-and-play functionality. The integration of Isaac Sim is a tool for development, not a turnkey solution. Buyers will still need to invest in AI expertise, either in-house or through integrators, to leverage these tools effectively. The source material does not disclose any specific service-level agreements, response times, or spare-part lead times for the new AI-enabled systems. These are critical operational factors that must be negotiated separately.
In the short term, the most practical takeaway is the availability of official 3D models for simulation. This is a tangible asset that can be used immediately for feasibility studies and virtual commissioning. For the medium term, the open-platform approach suggests that future robot controllers will be more like standard computers, capable of running a variety of AI workloads. For the long term, the influx of humanoid robots and advanced tactile sensing from China indicates that the definition of an "industrial robot" is expanding.
The iREX 2025 show was not about a single product launch. It was about the confirmation of a trend. The industry is moving from a model where the robot is a pre-programmed tool to a model where the robot is a perceptive agent, capable of understanding and reacting to its environment. The announcement from Japan’s largest manufacturer is the most authoritative validation of this trend to date. The presence of AgiBot and other Chinese firms is the competitive catalyst that will accelerate its adoption. For buyers, the message is clear: the future of automation is not just about buying a better arm; it is about buying a better brain, and the hardware is finally ready to accept it.
Published by Robot Service Map.