Inside Advanced Vision Solutions Co.: The Mitsubishi Sony AI Vision Sensor Venture

flowchart LR
    subgraph joint_venture ["Joint Venture"]
        A["Mitsubishi Electric"]:::blueNode --> C["Advanced Vision Solutions Co."]:::greenNode
        B["Sony Semiconductor"]:::redNode --> C
    end
    C --> D["AI Vision Sensors"]:::blueNode
    D --> E["Smart Manufacturing"]:::greenNode

    classDef blueNode fill:#2563eb,stroke:#fff,stroke-width:2px,color:#fff
    classDef redNode fill:#dc2626,stroke:#fff,stroke-width:2px,color:#fff
    classDef greenNode fill:#16a34a,stroke:#fff,stroke-width:2px,color:#fff

The manufacturing floor has always struggled with the bottleneck of visual inspection. Traditional camera systems flood networks with raw image data, overwhelming Programmable Logic Controllers (PLCs) and server infrastructure with severe latency and bandwidth issues. Now, a strategic partnership is shifting the paradigm. In July 2026, Mitsubishi Electric and Sony Semiconductor Solutions announced a joint venture named Advanced Vision Solutions Co., Ltd., fundamentally changing how we process visual data at the edge.

The Bottleneck of Traditional Machine Vision

For years, automation engineers have dealt with the immense complexities of integrating gigabit vision systems into industrial networks. When a high-speed assembly line produces parts at a rate of 100 units per second, capturing and transmitting high-resolution images to a centralized server for AI inference introduces unacceptable delays. The latency of data transit often forces machines to pause momentarily, or worse, allows defective parts to bypass rejection mechanisms before the control system receives the signal to actuate a pneumatic rejector cylinder.

The Mitsubishi Sony AI Vision Sensor initiative directly addresses this latency crisis by moving the neural network processing directly onto the image sensor itself, drastically reducing the physical distance that raw data must travel before a decision is rendered.

Why Edge AI Changes the Architecture

By embedding the AI processing logic within the sensor hardware, the system no longer outputs raw megapixel images to the network. Instead, it outputs lean, highly structured metadata: a simple boolean value (pass or fail), a specific defect classification code, or precise geometric coordinate data. This reduces the network load from gigabits per second to mere bytes per second. For SCADA networks and PLC architectures, this means vision data can be treated with the exact same determinism and low overhead as a standard inductive or photoelectric sensor.

flowchart TD
    subgraph traditional_vision ["Traditional Vision Architecture"]
        T1["Camera"]:::blueNode -->|Raw Images| T2["Industrial Network"]:::redNode
        T2 --> T3["Central AI Server"]:::redNode
        T3 -->|Pass/Fail| T4["PLC"]:::greenNode
    end

    subgraph edge_ai_vision ["Mitsubishi Sony AI Vision Sensor"]
        E1["AI Vision Sensor"]:::greenNode -->|Pass/Fail Metadata| E2["PLC"]:::greenNode
    end

    classDef blueNode fill:#2563eb,stroke:#fff,stroke-width:2px,color:#fff
    classDef redNode fill:#dc2626,stroke:#fff,stroke-width:2px,color:#fff
    classDef greenNode fill:#16a34a,stroke:#fff,stroke-width:2px,color:#fff

Inside Advanced Vision Solutions Co., Ltd.

The joint venture between Mitsubishi Electric and Sony Semiconductor Solutions, slated to begin operations in October 2026, combines Sony’s unparalleled expertise in CMOS image sensing with Mitsubishi’s deep integration in factory automation and control systems.

The partnership aims to develop specialized hardware that bridges the gap between IT (Information Technology) and OT (Operational Technology). Sony brings its IMX500 series architecture, which integrates logic and pixel layers to perform machine learning tasks on-chip at the moment of capture. Mitsubishi provides the critical OT context: robust industrial protocols (like CC-Link IE TSN), ruggedized hardware designs capable of withstanding extreme plant floor vibrations, and direct integration paths into the MELSEC PLC ecosystem.

This fusion of technologies resolves a major pain point for integrators who historically had to act as the middleman between fragile, IT-centric AI cameras and deterministic, rugged OT controllers.

Key Advantages for Automation Engineers

• Reduced Network Latency: Because image processing happens on the sensor, only structured metadata is transmitted to the PLC over the fieldbus, eliminating network congestion.
• Enhanced Privacy and Security: Raw images of proprietary manufacturing processes never leave the physical sensor housing, significantly minimizing the risk of data interception or unauthorized access.
• Simplified PLC Logic: Engineers no longer need to parse complex data streams or manage intricate handshakes with PC-based vision controllers. The PLC logic can be simplified to basic conditional statements driven by the sensor’s metadata output.
• Hardware Consolidation: This architecture completely eliminates the need for expensive Industrial PCs (IPCs) or dedicated vision controllers previously required to run heavy AI inference on the plant floor.

Comparing Vision Architectures

To fully understand the leap forward, consider the architectural differences between a standard GigE Vision setup and the new edge-native AI approach being pioneered by the joint venture:

The Future of Smart Manufacturing Integration

The formation of Advanced Vision Solutions Co. signals a broader trend in the industrial automation industry: the relentless decentralization of intelligence. As we push toward more autonomous and lights-out factories, the sensors themselves must become smarter. We are moving rapidly away from dumb sensors that require complex backend logic, toward smart, autonomous devices that hand off clean, processed decisions directly to the PLC.

For systems integrators, control engineers, and maintenance technicians, this evolution means less time wrestling with network bandwidth allocations, fewer points of failure in the vision system chain, and more time optimizing the actual machine control logic and overall throughput performance.

If you are looking to optimize your automation sequences, standardize your logic, or visualize this new metadata effectively in your SCADA environments, explore our AutomationView Store for advanced HMI templates and PLC simulation tools that can help you integrate edge AI directly into your next large-scale project.

Frequently Asked Questions

How does the Mitsubishi Sony AI Vision Sensor handle new AI models?

The smart sensors are designed to allow engineers to flash new, trained neural network models directly onto the device’s logic layer, meaning the hardware can dynamically adapt to new inspection tasks without needing physical replacement or rewiring.

Will this integrate directly with third-party PLCs like Siemens or Rockwell?

While Mitsubishi will naturally optimize for their own MELSEC ecosystem, the output of these smart sensors typically relies on standard industrial protocols like PROFINET, EtherNet/IP, or OPC UA, making them highly adaptable across different PLC vendor platforms.

What is the primary use case for on-chip AI vision?

The technology shines in high-speed visual inspection, surface defect detection, predictive maintenance monitoring, and rapid object classification where low latency, high determinism, and minimal network impact are strictly required.