Industrial embodied intelligence at scale just passed a key threshold. Drawing on nearly two decades of manufacturing-scene accumulation, Topstar and Wuwen Zhike built a shared data ecosystem and a complete industrial-data iteration loop, enabling fast iteration, standardised reuse and batch deployment of embodied robots.
Wuwen Zhike was among the first to launch a physical-AI data-base platform, covering AI large models, industrial robots and embodied intelligence, and has won tens of millions of yuan in simulation-test orders from leading firms.
Topstar and Wuwen Zhike built the first industrial-handling physical-AI data infrastructure loop, covering virtual simulation, data labelling, data sorting, model training and real-machine evaluation, giving Topstar’s embodied products core support for continuous evolution.
Virtual simulation layer compresses debugging cycles. A self-built physics engine builds high-fidelity digital twins. Scan2Sim and Gen2Sim batch-generate multi-condition training scenes and synthesise long-tail edge cases, cutting on-site debugging from months to weeks.
Data labelling layer lifts efficiency tenfold. Simulation data carries perfect ground truth. real data uses auto pre-labelling plus human review, raising efficiency ten times.
Data sorting layer builds a perpetual positive data flywheel. A data highway unifies simulation, real-machine and production-line sources, with automated cleaning, layering and versioning, while real-machine failures automatically flow back to simulation for amplification, forming a find problem, add data, upgrade model loop.
Model training layer enables continuous self-evolution. A perception, planning, control three-tier framework, with simulation pre-training plus real-machine fine-tuning and support for incremental learning and hot updates, lets ability grow with operation.
Real-machine evaluation layer quantifies ability. An embodied simulation-evaluation framework, with metrics for task success rate, efficiency and safety, gives high-fidelity quantitative assessment.
On this virtual-real dual base, Topstar has connected data capture, simulation expansion, model training and real deployment. Its full embodied-robot line iterates on the shared platform, free of single-scene custom R&D, enabling fast reuse of technical solutions and cross-industry batch copying, opening long-term growth for multi-scene commercial use.
Topstar pursues its mission to make industrial manufacturing better, using embodied intelligence, industrial robots, CNC machines and injection moulding machines as four core smart devices, plus control, servo and vision as three core technologies, to build an AI-driven smart-hardware platform aimed at becoming a global leader in embodied intelligence.
Images
Editor’s note: This is an adapted translation of the original OFweek report. It has been trimmed and restructured for readability for an international business audience.