“ In the second half of the game, the player is more intelligent than the depth of the landing.
From October 12 to 16, the 26th China International Industrial Exhibition will be held at the National Exhibition Centre (Shanghai). Agile Robots will present 12 exhibitions in the five main exhibition areas: "Intelligent, Physical Model, Research Ecology, Consumer Applications, Industrial Landscape", which will showcase all the technical capabilities from humanoid robots, models, force control and sensing technologies, data collection systems, to multi-scenario solutions.01.
Distance from the model to landing
Over the past two years, physical intelligence has gone through a complete cycle of explosion from concept to rational regression. The heat of capital has spawned a large number of man-made integrations, but the market has quickly discovered: form is not equal to capacity, demonstration is not equal to delivery. The reason is not complicated: models can generate a reasonable action plan, but friction in contact, object deformation, deviation in the position of the workpiece, still need to be sensed by a system, real-time control and co-operation with the executive.
The company's advantages are in the long-term investment and development of this technological chain. From motors, decelerators, sensors, to mechanical arms, shrewd hands and humanoid robots, to robotic operating systems, basic models and application algorithms, it has built a full-scale capability to cover core software hardware and extend it to production lines and factory applications.
It's not about putting more technology into a device, it's about optimizing collaboration around specific tasks: making the task of understanding a model into a hardware executable action, letting access to feedback into a control loop in a timely manner, and letting on-the-spot problems in turn drive product transformation. As the founder and CEO of Intelligent Robotics, Dr. Chen Zheng, says: "Physical intelligence is not just about embedding big models into a robot body. It requires model, sensor, executive system, and application landscape collaboration, rather it requires large-scale deployments through high-price applications, forming a continuous cycle of 'apply-to-apply-to-apply-to-apply-to-apply-to-apply-to-apply-to-apply-to-apply-to-apply-to-apply-to-apply' data".
02.
How to make physical AI real
Mobile FR3 Duo: Installing a "mobile body" for the study of physical AI
As the study of physical intelligence moves from desktop operations to dynamic space, mobility, perception and arm operations are no longer interdependent topics. The collaboration between the dashboard and the mechanical arm, the integration of sensory data, the execution of strategies on a real machine, may become an engineering burden before algorithmic validation.
Franka Robotics' Mobile FR3 Duo arm mobile research platform, which integrates the FR3 Duo benchmark platform with touch mobile robots (TMR), integrates arm operations, multi-mode sensing and all-in-one mobility to enable remote operations, data integration and strategy execution on the same platform.
In 2026, the first EBiM (Embodied Bimanual Manipulation) Challenge officially kicked off, attracting 119 teams from around the world to sign up. The event aims to build a physically intelligent standardized assessment system geared towards the real world, allowing robot algorithms not only to be effective in simulation, but also to be validated in different laboratories, environments and on real robot platforms. As the main sponsor of the event, Intel offers a unified platform called Mobile FR3 Duo to support global researchers in completing real robot tests and allowing algorithms to be tested on real hardware.
For research teams seeking real-world validation, the value of Mobile FR3 Duo's mobile arm research platform not only increases mobility, but also allows algorithms to enter the physical world, shortening the distance from research imagining to real-machine validation.
2026 is the starting point, and 2027 will be further expanded. EBiM 2027 will enter a new Beta phase, with plans to further expand the entry teams, task types and test platforms, and continue to refine the standardized evaluation protocols. We invite universities, research institutions, developers and robotic enterprises from around the world to join us in creating algorithms for authentication in real robots, real missions and real environments.
AI Therapeutic Robots: Fireworks of Physical AI
At this year's trade fair, the AI therapy and rehabilitation robot was the easiest to experience at the SILING platform. This is a reflection of the attractiveness of SILING's excellent products, while also representing real market demand.
As the healthcare industry enters a period of rapid growth, with a profound shift in overlapping demographics, the demand for therapeutics continues to unfold, but the supply-side contradictions are also sharper: stores are plagued by difficulties in recruiting professional technicians, low retention rates, difficulty in unifying service standards, high labor costs, and consumer demands for safety, personalization, and immersive experiences are rising.
The solution to this structural contradiction is to bring physical AI into the rehabilitation scenario. The AI rehabilitation robot is equipped with a self-developed high-precision power control mechanical arm, equipped with a five-terminal acceleration system, a device that can switch seamlessly between the five specialized therapeutic modes of internal heat, massage, sleep regulation, dot wave and electrical power. The method of massage relies on high-precision power control technology, strength stability, programmable trajectory, precision-momenting techniques, which provide consistent positioning for problems such as muscle fatigue and physical imbalance.
For operators, it is also a set of digital operating systems: the Cloud 100+ standard curriculum system allows new technicians to work half a day, translate their credential experience, hands-on services, into standard processes for replicable and quantifiable filters; supports automated robotic work + artificial hand-held filters that work in parallel, reshaping filtering efficiency and human efficiency; and the full-link management system automatically absorbs equipment using data and member health profiles, enabling store operations to move from filtering sensors to data-driven filters.
Photographic robots: freeing creative freedom
The boom in film industrialization and short video and live broadcast e-commerce is pushing the demand for professional photography to an unprecedented level, while high-quality cameras have long depended on two scarce resources: expensive rocking arms, sleighs, electrically controlled cloud platforms, and experienced photographers.
It combines AI, power control, and photography workflow, allowing photographers to use robots with simple, familiar creative logic. Unlike the complex demonstration operations of industrial robots, it supports drag-and-drop, allowing users to directly guide the machine arm to the target position, set a critical axis, and generate a mirror path using a trajectory-planning model.
Around the image control, the system is equipped with a focus modulator, which supports custom-defined key image focusing and automatically decomposes the focus information during the shooting process, completing the tracking. The lens and focus control are interconnected, allowing the robot to not only repeat the movements, but also perform more performance shooting design.
Repeatability is further translated into production efficiency. The shooting trajectory can be saved, shared and replicated, providing support for the shooting of the same goods, multiple rounds of creative adjustment and modeling production. Game handle control, sound cue and normal 220V power supply, further reducing equipment use thresholds; easy to transport structural design, also in line with the transfer needs of commercial shooting.
The ultimate value of AI photography robots is not that they make judgments instead of photographers, but that they put complex, repetitive execution in the hands of robots, allowing creators to put more effort into designing, narrating and expressing. This is similar in underlying logic to the 100+ curriculum system of AI therapeutic robots, which converts rare techniques that rely on talent and accumulation into storable, replicable, and distributed digital assets.
This is perhaps the most important value of physical AI to the millennial industry: it not only replaces repetitive labor, but also liberates human experience from the individual into a scalable and distributed productivity.
03.
Full-fledged, stereo presentation
In addition to the three main exhibitions, the rest of the exhibitions further demonstrate the technical depth of thought from industrial, scientific and ecological dimensions.
The Diana 3 Duo automotive parts line brazing solution faces typical difficulties in flexible manufacturing: the line brazing is deformable, the state is unstable, it is difficult to perform steady operation based on a pre-set trajectory. The solution is combined with the self-developed seven-axis controlled mechanical arm Diana 3, designed for the flexible brazing/line-line/line brazing process for automotive parts (such as field controllers, seat brazing), which relies on power control + vision + Agile Core system to achieve line brazing and assembly.
Diana 3, industrial-grade power control, 3 kg load, entry-level, lightweight scientific research experimental scenarios, retaining 7 degrees redundancy structure, joint torque sensing and smooth control, and sharing the same set of development interfaces with the Diana 7, FR3 also support FCI. Precision operations research, laboratory automation, medical research aids, sample processing and small replication operations are its main battlefields.
Agile Core+ visual touch sorting combines sensing and system integration capabilities. The VT10-G visual touch sensor can output RGB images, dynamic field distributions, depth charts and three-dimensional clouds to capture contact information; Agile Core is a self-developed robotic operating system and software platform that is positioned to serve as a robot's mastered smart brains, unified connecting robots, visual power controllers, mobile robots, and PLCs, as well as a gateway to process development and cross-device deployment through drag-and-drop programming and interface expansion.
The high-precision power feedback remote operating system is based on the new generation Diana 7 from the main remote operating platform. The new generation Diana 7 is upgraded with more open control and more flexible applications. It can achieve multi-second bi-directional power feedback, full-junction torque sensing, real-time touch-to-two-direction transmission, operators can remotely sense touch, and shares the same set of development to interface packs with Diana 3 and FR3 as well as support for FCI. From laboratory algorithm research, real-world system verification, the new Diana 7 is dedicated to becoming a robotic platform that connects Physical AI innovation and industrial applications.
Anti-interference demonstrations based on the WAM model rely on the FR3 Duo hardware platform and are based on the FastWAM optimization, which greatly enhances the robustness of environmental visual interference and explores the model's ability to execute reliably under changing conditions. The agile hand focuses on the execution and perception of precise operations. Agile Hand uses a modular five-finger design, combined with multi-sensory systems and dynamic smooth controls, that can be extended to 16 degrees of freedom by upgrading the anti-finger executor.
The three Franka collaborative ecological exhibitions combined the open platform with partner capabilities: two-arm collaboration SMT multi-step precision operation, two-arms FR3 featuring all-natural TeleHand Lite handhelds, simulating real factory quality inspection, displaying grabbing, flipping, visual verification and touch detection, and other quality inspection processes; wearable inertial motion capture synchronous systems, FR3 combined with j-mex AgileMaster, can be refined to simulate human posture, achieve high synchronous asynchronous effects, open data collection, data transmission and robot replication, efficient types of robots, remote human-operated and human-operated remote controls; user interaction with drives, DuoRobospec, DuoRobospec, integrated with real-world displays, and the ability to restore real-world visibility, v5.
The Agile Robots presented at this year's trade fair are not a bunch of isolated products, but a complete set of industrialization capabilities: continuous innovative R&D systems, AI and robotics technologies geared toward the physical world, scalable manufacturing capabilities, and delivery networks that connect global markets.
Since its founding in 2018, Agile Robots has built a fully self-developed and fully chain-controlled system around WAM (world action model), perception, decision making, power control, vision and operating systems. Today, Agile Robots has grown into a global enterprise with more than 3,500 employees from more than 60 countries, headquartered in Munich, Germany, with annual revenues doubling for eight consecutive years and delivering nearly 40,000 robots to customers worldwide.
Agile Robots, a mind-blowing robot based on full-featured technology, is pushing physical AI to every corner of the world based on customer value.