Luming Robotics Showcases NexCore at the 88th China Educational Equipment Exhibition, Demonstrating a Closed-Loop Embodied Intelligence Training Program

Luming Robotics' Full Product Line

On October 30, the 88th China Educational Equipment Exhibition opened at the Changsha International Convention and Exhibition Center. As soon as the exhibition opened, the Luming booth W4003 drew layer upon layer of crowds: the AI mini bipedal humanoid robot Lu Xiaoming danced to dynamic street dance music, the Lumos MOS2, known as the "heavy-duty AI Worker," skillfully picked up turnover boxes and stacked them in place, and the Lumos Station Lite skill development system demonstrated on-site skills such as folding clothes, flower arranging, and operating a toaster.

Beyond the bustle, a large screen in the center of the booth scrolled through task flows and evaluation metrics, which made many school leaders and program heads stop to look closely. It is reported that what was running on the screen was Luming's self-developed robot skill evolution engine—Lumos NexCore. From data collection and model training to real-machine deployment and quantitative evaluation, every stage was marked with its status and results. A visiting teacher asked on the spot: "Can this closed loop be directly broken down into a comprehensive practical training course?"

At the exhibition site, visitors used Luming's master-slave arm teleoperation to control the robotic arm to grasp objects and personally experience "real-machine practical training."

This question pinpointed the difficulty currently facing the development of embodied intelligence programs in universities and vocational colleges.

Industry is scaling up deployment, while classrooms are stuck on scenarios, data, and teachers

Embodied intelligence is entering a period of large-scale deployment. In the first half of 2026, financing in China's embodied intelligence sector reached 93.5 billion yuan, an increase of about five times year on year; domestic humanoid robot shipments exceeded 40,000 units, accounting for about 97% of the global total.

The industry's boom cannot be separated from the accelerating boost of local policies. Taking Hunan as an example, the implementation opinions issued by the provincial government office set out a clear timetable: by 2028, the province's intelligent robot industry chain revenue is to strive to exceed 100 billion yuan, cultivating about 5 chain-leading enterprises; by 2030, it is to exceed 200 billion yuan, with the industry's comprehensive strength entering the country's first tier.

Facing the huge talent gap on the industry side, the education side's program catalog is also rapidly following up. In 2025, the national vocational education program catalog added "Embodied Intelligent Robot Technology" and "Embodied Intelligent Engineering Technology," and in 2026 it further added programs such as "Humanoid Robot Engineering Technology," "Embodied Intelligent Computing Technology," and "Marine Intelligent Robot Application Technology."

Industry is advancing triumphantly, policies continue to intensify, and the education side has also "launched new offerings" accordingly. But to truly bridge industry-education integration, updating program names alone is far from enough. The most realistic difficulties facing institutions remain those three things: teachable scenarios, usable data, and capable teachers.

Specifically in the construction of training rooms, many institutions easily fall into three pitfalls: emphasizing equipment over scenarios, buying all the equipment but having no tasks to run, so the training room ultimately becomes a showroom; emphasizing simulation over real machines, performing beautifully in simulation but failing as soon as a real machine is used; emphasizing algorithms over engineering, turning courses into "shrunken graduate courses," so students do not learn deeply and cannot use it on production lines.

The key prerequisite for solving these problems is to truly bring the industry side's real tasks, data standards, and acceptance logic into the classroom.

At the exhibition site, Luming demonstrated one of the core functions of an industrial embodied intelligence training room: master-slave arm teleoperation hovering

NexCore compresses the industrial task closed loop into the training room

Yu Chao, founder and CEO of Luming Robotics, proposed that as embodied intelligence enters its second half, the core of competition is shifting to the task closed loop.

Supporting this judgment is Luming's self-developed NexCore robot skill evolution engine. It links data acquisition, model training, skill evaluation, and deployment operation into a reusable production line, enabling "data—model—skill—execution—feedback" to form a closed loop.

Luming's industry-first full-process industrial embodied skill validation NexCore

In reports from the site of the 2026 World Robot Conference, both CCTV and Xinhua mentioned that this system can compress the cycle for a robot to learn a new skill from several weeks to several days; after skill deployment, real operation data flows back for retraining and can be reused across more robots and scenarios.

CCTV1 focused on Luming MOS's real handling and quality inspection operations, with the entire process driven by the NexCore system

For institutions, the significance of NexCore lies in being "teachable."

Taking the MOS robot at this exhibition as an example, students can complete data collection through teleoperation, use the collected real data to train a box-handling model, and then deploy it on a real machine for execution and continuous iterative optimization.

Luming is precisely transforming such real industrial site tasks into teaching projects, forming six major training room directions covering "classroom teaching—project practical training—model validation—competition training—industrial application": joint modules and drive control, robotic arm control and intelligent operation, industrial embodied intelligence applications, humanoid robot systems and applications, multi-source embodied data collection and processing, and robot repair and maintenance.

With NexCore as the foundation, Luming transforms real industrial tasks into a teachable and assessable practical training system

The curriculum system progresses through four levels: L1 foundation builds a solid grounding in Linux, Python, ROS/ROS2, and robotic arm kinematics; L2 perception and control completes camera calibration, visual recognition, drag teaching, and trajectory planning; L3 embodied data and models moves into teleoperation data collection, data annotation and quality inspection, imitation learning, and VLA deployment; L4 comprehensive scenario projects directly target real tasks such as flexible quality inspection, sorting, and handling and palletizing, concluding with metrics such as task success rate, execution cycle time, exception recovery, and data quality.

Taking flexible quality inspection as an example, a practical training project can be broken down into five tasks: workstation setup and calibration, teaching and pick-and-place, visual inspection and judgment, data collection and training, and integrated debugging and quantitative evaluation. These cover the four stages of "collection, tuning, testing, and use," and the assessment metrics are the same as enterprises' acceptance standards.

The training room is configured by scenario: desktop workstations for practicing basic single-arm and dual-arm operations; data collection workstations using VR teleoperation and handheld data-collection grippers to practice "accurate collection"; and scenario production-line workstations using wheeled-arm dual-arm robots to connect with real processes such as quality inspection, assembly, and handling.

The positioning of institutions also becomes clear accordingly: research-oriented universities focus on new algorithms, new models, and new embodiments, while vocational colleges and application-oriented universities train "collection, tuning, testing, maintenance, and use" into students' muscle memory.

Why Luming: an industrial foundation underpins its education solutions

Luming enters the education sector bringing task definitions, data standards, and acceptance logic that are already in use on the industrial side.

At the model layer, Luming's self-developed Prime R0, with 2.8B parameters, achieved the world's No. 1 overall success rate on the MolmoSpaces zero-shot embodied benchmark and can run locally on consumer-grade graphics cards, lowering the computing-power threshold for institutions.

At the embodiment layer, Luming's MOS series heavy-duty AI Workers are already performing tasks such as bin handling, flexible quality inspection, and intelligent assembly in real factories; the bipedal mini humanoid robot Lu Xiaoming won two golds and one silver at the Second World Humanoid Robot Games, ranking among the top four in the industry.

Luming MOS2.0 performs a material handling task on site

At the industrial ecosystem layer, Luming transforms the real scenarios of industrial partners such as JD.com, Mitsubishi Electric Intelligent Manufacturing (China), and COSCO Shipping into teaching tasks; meanwhile, Luming has established industry-academia-research cooperation with nearly 10 institutions and is carrying out in-depth coordination with schools on jointly building standardized embodied intelligence training rooms and jointly developing industrial teaching sandboxes and course resources.

Faculty and evaluation mechanisms are keeping pace.

Teachers are stationed on site in enterprise projects and bring back real working conditions; enterprise engineers enter the classroom and take on comprehensive project courses; both sides jointly develop project packages and new-form teaching materials, and apply for research topics using enterprise projects as carriers. Evaluation shifts from reports to metrics: task success rate, execution cycle time, exception recovery, and data quality are automatically recorded by the process, and the results can be compared horizontally and benchmarked against enterprise acceptance.

Relying on four core industrial scenarios, Luming transforms real production-line tasks into teachable and verifiable teaching projects

One real industrial problem can grow three things: teaching resources, research topics, and accumulated outcomes; in terms of collaboration mechanisms, this means jointly building scenarios, jointly collecting data, jointly researching curricula, and jointly cultivating talent.

See the real machines on the first day, see the solutions on the 31st

On the opening day, the MOS2 real-machine operation, the Lumos Station Lite multi-skill demonstration, and the Lu Xiaoming dance performance at Booth W4003 continued to attract visitors.

For institutions that are applying for embodied intelligence-related majors and planning phased construction of training rooms, what is more worth a closer look is the task loop behind NexCore and how the six major training room directions connect with the curriculum system and job competencies.

From 14:00 to 17:00 on October 31, the "Integration of Education and Science, Symbiosis of Industry and Education" Embodied Intelligence Education Solution Sharing Session and New Product Launch will be held in Meeting Room 2008, Second Floor, North Registration Hall, Changsha International Convention and Exhibition Center.

Luming will debut a master-slave arm teaching and demonstration platform, directly targeting the most fundamental data collection step in embodied intelligence education. Zhao Xuzuo, Technical Director of the Flying Car Research Institute of Tsinghua University Suzhou Automotive Research Institute, will give a keynote sharing on "how real industrial needs can be transformed into teaching resources and research topics."

University perspectives, Luming's solutions, new product launches, ecosystem practices, and exchange and matchmaking will unfold in sequence, providing institutions with a transformation path from industrial tasks to course projects and then to research topics.

What is scarcest in embodied intelligence education is a task chain that teachers can explain clearly and students can run through completely. Luming has brought this chain from the industrial front line to the site of the China Educational Equipment Exhibition.