How did the Humanoid Robot Innovation Center in the high-end manufacturing industry implement an input invoice management platform?

Make recognition and verification an automatic step in the business process, completing data retrieval, authenticity verification, duplicate checking, and compliance validation upon invoice submission; abnormal invoices are immediately alerted and returned to the handler, no longer entering subsequent processes; recognition results and verification records are retained together, forming traceable processing records.

What problems existed at the Humanoid Robot Innovation Center before launching input and output invoice management?

Procurement in the manufacturing stage is spread across raw materials, parts, molds and tooling, equipment spare parts, outsourced processing, and logistics. Suppliers include both large manufacturers with long-term agreements and small and micro suppliers providing temporary replenishment, with varying invoice types and arrival rhythms; invoice images are photographed and uploaded by the handler, and shooting angle, lighting, and wrinkles all affect readability, so finance often has to recheck against the original when reviewing; under spot-check methods, abnormal invoices are often only exposed at the certification stage, by which time the business has already completed payment.

← Back to case list

High-end manufacturingInput and Output Management

Humanoid Robot Innovation Center · National-level humanoid robot innovation platform

Humanoid Robot Innovation Center | Input VAT invoice pool + OCR verification

National-level platformEnterprise Attributes

OCR + invoice poolCapability Scope

Enterprise overview

Beijing Humanoid Robot Innovation Center Co., Ltd. was jointly established at the end of 2023 by ten leading industry enterprises and institutions including UBTECH, Beijing Jingcheng Machinery Electric, Xiaomi Robotics, and E-Town Robotics. In October 2024, it was officially inaugurated as the "National and Local Jointly Built Embodied Intelligent Robot Innovation Center." In April 2024, the center released the world's first full-size humanoid robot with pure electric drive and human-like running, "Tiangong," standing 163 cm tall and weighing 43 kg, equipped with multiple visual perception sensors, a high-precision inertial measurement unit, 3D vision sensors, and six-axis force sensors. It adopts a self-developed method of "predictive reinforcement imitation learning based on state memory" and can run stably at 6 km/h; the new-generation "Tiangong 2.0" further achieves free walking on complex terrain, ultra-long battery life, and battery swapping while standing.

Business pain points

Procurement in the manufacturing stage is spread across raw materials, parts, molds and tooling, equipment spare parts, outsourced processing, and logistics. Suppliers include both large manufacturers with long-term agreements and small and micro suppliers providing temporary replenishment, with varying invoice types and arrival rhythms. Invoice images are photographed and uploaded by the handler; shooting angle, lighting, and wrinkles all affect readability, and finance often has to recheck against the original when reviewing. Under spot-check methods, abnormal invoices are often only exposed at the certification stage, by which time the business has already completed payment.
Voided invoices and red-flushed invoices mixed into accounts directly affect input VAT deduction and cost recognition. Entry errors will propagate along authentication, deduction, and vouchers, making post-correction costly
Duplicate reimbursement relies on memory and ledger comparison, and duplicates across periods and departments are hard to detect. After China Golden Tax Phase IV, tax comparison tends toward full quantification; once invoice authenticity has problems, it directly affects deductions, and spot-check verification can no longer cover regulatory requirements

Solutions

Make recognition and verification an automatic step in the business process, completing data retrieval, authenticity verification, duplicate checking, and compliance validation upon invoice submission
Anomalous invoices trigger immediate alerts and are returned to the handler, no longer entering subsequent processes
Recognition results are retained together with verification records, forming traceable handling traces
Ultimately shifts finance from checking one by one to handling only exceptions, improving both invoice processing efficiency and accuracy
At the same time, complete recognition and verification traces are retained, providing a basis for subsequent audit tracing and error analysis