What Is Lumos NexCore
Lumos NexCore is an embodied intelligence evolution engine launched by Lumos Robot, positioned as skill infrastructure for industrial deployment. Lumos NexCore integrates data assets, model training, evaluation and validation, skill management, and robot operations into platform services, transforming robot skill development from a手工 workshop-style process into assets that can be trained, reused, and continuously iterated, enabling them to be invoked and validated like cloud services, and driving embodied intelligence from single-point demos toward large-scale industrial applications.
Main Features of Lumos NexCore
- Data Asset Platform: uniformly aggregates real, simulated, and video data, making every piece of data searchable, annotatable, and reusable, solving the raw material problem of skill production.
- Model Training Platform: hosts task orchestration, parameter optimization, and process tracking across the entire training workflow, allowing customers to launch model training with one click without building their own algorithm environments.
- Evaluation and Validation: through multidimensional evaluation and capability diagnosis, model performance is transformed into quantifiable and reproducible acceptance evidence, giving skill quality a reliable basis.
- Skill Management: packages trained models together with code, parameters, and execution strategies into skill units that can be published, versioned, and reused across scenarios.
- Robot Operations: connects real devices and automatically feeds operating status, logs, and high-value samples back to the platform, driving the next round of training and skill iteration.
- Natural Language Orchestration: users only need to describe task goals in spoken language, and the platform can automatically orchestrate the subsequent data preparation, training, and deployment processes.
Technical Principles of Lumos NexCore
- Closed-Loop Flywheel Mechanism: real robots execute tasks → generate new data and failure samples → data flows back to the platform → retrain and optimize models → generate new versions of skills → deploy back to devices for execution. Through a continuous data closed loop, robot capabilities self-evolve in real operations.
- Platformization Replacing Project-Based Development: traditional robot skill development is "workshop-style": every new scenario requires repeating collection, training, parameter tuning, deployment, and rework. NexCore standardizes and manages this process, turning skill production from reliance on engineers' individual experience into observable and traceable platform services.
- Skill Assetization: models are encapsulated as "skill units" containing code, parameters, and execution strategies, enabling cross-scenario reuse, version management, and continuous iteration, and making them invokable like cloud services.
- Effective Data Scaling: unlike simply pursuing data collection duration, NexCore emphasizes effective data volume—that is, data that covers the task distribution, includes failure and recovery samples, can transfer across robots, and can truly improve model success rates. Its underlying Scaling Law is built on "task experience" rather than "data volume."
- Form Decoupling: the platform is not bound to a specific embodiment, and the core technology stack can be transferred and reused. Hardware form is determined by task constraints, while the brain (the model and skill system) remains unified, enabling capability reuse across scenarios.
How to Use Lumos NexCore
Send an email to builder@lumosbot.tech to apply for a trial. Describe task goals in natural language, and the platform can automatically complete the entire process from data preparation, model training, and evaluation and validation to multi-machine deployment, continuously iterating and optimizing skills during real operation.
Core Advantages of Lumos NexCore
-
Closed-Loop Evolution: Connects the entire chain of data, training, evaluation, skills, and device operations, enabling robots to continuously self-iterate in real-world operations.
-
Low-Barrier Development: Supports natural language descriptions of task objectives and automatically orchestrates subsequent processes, allowing business-savvy people to participate in skill development without an algorithm team.
-
Skill Assetization: Packages models into publishable, reusable, and version-manageable skill units, realizing the transformation from project customization to platform-based invocation.
-
Hardware Decoupling: Compatible with multiple robot bodies in both the Luming ecosystem and non-Luming ecosystems, skills can be trained and used immediately without being limited by hardware form.
-
Industry Validation: Tempered on real production lines such as Mitsubishi Electric, its flagship brain Prime R0 ranks first globally in the MolmoSpaces zero-shot evaluation.
-
Effective Data Scaling: Drives evolution through task experience rather than mere data volume, emphasizing the real value of data that covers task distribution, failure samples, and cross-machine transfer.
Comparison of Lumos NexCore with Similar Competitors
| Comparison Dimension | Lumos NexCore | NOKOV ShadowEngine |
|---|---|---|
| Product Positioning | A "skill infrastructure" for industrial embodied intelligence, allowing robot capabilities to be invoked, trained, and validated like cloud services | An end-to-end robot AI training platform that integrates the full chain of motion capture, data management, retargeting, teleoperation, training, and validation |
| Core Architecture | Five major modules: data assets, model training, evaluation and validation, skill management, and robot operations | Six major modules: ShadowEngine Hui, ShadowEngine Yuan, ShadowEngine Yi, ShadowEngine Yao, ShadowEngine Xun, and ShadowEngine Yan |
| Data Closed Loop | Supports unified indexing of real, simulated, and video data, with real-machine operation data automatically flowing back to drive iteration | Real-time aggregation and synchronization of multimodal data, Sim2Real closed-loop validation, and real-machine data feedback to the training platform |
| Skill Reuse | Packages models into publishable, reusable, and version-manageable skill units, supporting cross-scenario invocation | Provides 1,000+ hours of standardized human motion data assets, supporting batch simulation and export of expert training data |
| Hardware Adaptation | Compatible with multiple robots in both the Luming ecosystem and non-Luming ecosystems, without binding to a specific brain or body | Already adapted to mainstream humanoid robots such as Unitree G1, supporting rapid access via URDF/MJCF open interfaces |
| Usage Barrier | Natural language description of task objectives automatically orchestrates the entire process, with no need to build an in-house algorithm team | Requires optical motion capture equipment and a simulation environment, targeting research teams and robotics companies |
Application Scenarios of Lumos NexCore
-
Scientific Research Experiments: Provides universities and laboratories with a full-chain research platform from data collection to skill deployment, accelerating the iteration of embodied intelligence algorithms.
-
Industrial Manufacturing: Completes tasks such as quality inspection, logistics, handling, and loading/unloading on factory production lines, and continuously optimizes skills through real operation data.
-
Commercial Services: Supports the full "pick-carry-deliver" service process in scenarios such as hotels and buildings, enabling rapid cross-task adaptation and deployment.
-
Cultural and Entertainment Display: Provide orchestration-capable, iterable skill assets for robot performances and exhibition interactions, supporting reuse across multiple hardware forms.