Mitsubishi Electric and Sony Semiconductor Solutions have agreed to establish a joint venture, "Advanced Vision Solutions," to develop AI vision sensors for the manufacturing industry. Subject to obtaining necessary regulatory approvals and other conditions, the companies plan to launch the business in October 2026. The aim is not simply to put AI on cameras, but to pass recognition results to manufacturing equipment for decision-making and control, creating a system usable on the factory floor. That said, October marks the start of company operations, and product release dates and specifications have not yet been announced.
A 60-40 split, with operations starting in October 2026
The new company will be 60% owned by Mitsubishi Electric and 40% by Sony Semiconductor Solutions, headquartered in Minato Mirai, Yokohama. Tamiki Kobayashi will serve as representative. The stated business scope explicitly includes the development of AI-equipped vision sensors for factory automation (FA) and related solutions.
The roles each company brings are also clear. Sony will provide image sensor and edge AI technology, while Mitsubishi Electric will combine its FA control expertise with its digital platform "Serendie." Given Mitsubishi Electric's majority stake and the business's focus on FA applications, the new company is positioned not to sell general-purpose camera components, but to develop products and systems that incorporate recognition results into equipment control.
The two companies state that they will capture equipment status and product quality, connecting the necessary information to decision-making and control. Operational status and maintenance information are also included as targets. They further envision a multimodal production system that combines visual data with other on-site data to detect changes and early signs of problems. Labor savings and unmanned operation are effects that would follow once such a system runs stably on-site—they are not achievements already demonstrated at the time of announcement.
The gap between in-sensor AI and FA control
Sony's Intelligent Vision Sensor stacks a pixel chip with a logic chip equipped with an AI engine and dedicated memory. Rather than continuously sending all captured images externally, it can output metadata indicating detected objects or select and crop only the necessary regions of an image. Sony explains that this configuration reduces data transfer volume and communication costs while also addressing latency and privacy concerns.
Even so, manufacturing equipment cannot operate on recognition results alone, even after a sensor identifies an object. A pathway is needed to convert recognition results into information that FA equipment can use, cross-reference it against the equipment's operating conditions, and reflect it in control commands. This joint venture aims to close the gap that has remained between in-sensor AI processing and the factory control loop.
Serendie is Mitsubishi Electric's platform for leveraging data across business domains to create services. According to official descriptions, it covers FA systems as one of its target domains, combining data gathered from inside and outside the company with technologies such as AI and robotics. The concept of using visual data alongside equipment operating history and maintenance information for the same decision-making overlaps with this platform's role. For a venture that champions physical AI, what distinguishes a product's real capability is not the ability to understand images, but how safely and promptly it can feed those results back to actual equipment.
MELSENSOR's discontinuation, and the options that follow
In May 2020, Sony commercialized the "IMX500," an image sensor equipped with AI processing capability. In 2021, it began offering "AITRIOS," an edge AI sensing platform that supports AI model development, device management, and application development. This announcement does not mark the starting point of in-sensor AI. Rather, it represents a move to create a business entity that brings existing technology to market integrated with manufacturing equipment control.
Mitsubishi Electric also has its own history in this area. The company's FA sensor line, "MELSENSOR," offered vision sensors that automated inspection and measurement, but the official product page lists the vision sensors and code readers as discontinued as of September 2025. There is no stated explanation that the AI vision sensors being developed by the new joint venture are a direct successor to MELSENSOR. Even so, it is significant that Mitsubishi Electric has called the AI vision sensor a "new core component" and indicated a policy of incorporating it into its FA digital solutions business.
| Timing | Sony/Mitsubishi Electric Development | Difference from This Announcement |
|---|---|---|
| 2020 | Sony commercializes IMX500 | AI processing executed on the sensor |
| 2021 | Sony launches AITRIOS | Provides a platform supporting development and operation |
| September 2025 | Mitsubishi Electric's MELSENSOR vision sensors discontinued | Legacy products discontinued; direct successor relationship not disclosed |
| Planned October 2026 | Advanced Vision Solutions begins operations | Joint business including FA decision-making and control |
What this timeline shows is that the novelty here lies not in the invention of AI image sensors, but in a structure that provides, as a package, everything from recognition to control at the manufacturing site. How much of the burden of integrating components, development platforms, and FA control—which would otherwise need to be procured and integrated separately—the new company takes on will determine the value it offers to customers.
What to watch to evaluate "decision-making and control"
The announcement does not specify the model numbers of the sensors to be adopted or the AI models involved. Recognition accuracy and processing time are also not disclosed. Beyond the compatible FA equipment and connection interfaces, pricing and release timing remain unannounced as well. While the existing IMX500 and AITRIOS can be referenced as underlying technologies, there is no guarantee that the new company's products will use the same configuration or performance.
For manufacturing sites to decide on adoption, they will need to verify recognition performance under changing lighting conditions and target objects, as well as the cycle time at which recognition results can be passed to control equipment. How AI models are updated for each site, whether images or metadata are stored, and what connection method links Serendie with existing equipment will also affect implementation costs. The product the two companies envision—one usable "without specialized knowledge of optics or AI"—will be judged by how far these settings and operations can be standardized.
The first milestone to confirm is whether the new company, after obtaining regulatory approvals, begins operations as planned in October 2026. After that, attention will turn to the disclosure of product model numbers, compatible equipment, and connection specifications. Once verification results are also available, it will become possible to judge how far the concept of handling everything from recognition to control under one company can go toward reducing labor needs and enabling autonomous improvement on factory floors.