Japan’s Factory Innovation Week 2026 exhibition showcased a range of AI-powered machine vision solutions, with a clear focus on using robots on production lines handling multiple products or varying quantities. The event was held from September 9 to 11, 2026, and included demonstrations in the “RoboDEX” pavilion focused on making robots more capable of understanding and handling objects.
Among the solutions on display was a system using a 3D vision camera mounted on the robot arm, along with AI models designed to recognize objects and determine how to pick them up. This approach targets tasks that are difficult to perform using fixed rules, such as picking up randomly placed parts or handling products that differ in shape and size.
From Recognition to Executing the Picking Operation
The demonstrations explained that combining 3D vision and artificial intelligence is not limited to determining the location of a part; it also enables the robot to estimate its orientation and shape before being directed. As a result, the robot can recognize irregular or mixed parts and then select the appropriate picking path without requiring complete reprogramming when the product changes.
The proposed solutions included software packages for tasks such as arranging products on pallets and depalletizing them. These applications enable the handling of situations in which the dimensions or arrangement of parts vary, while reducing the need to manually adjust settings each time the product’s shape changes.
Solutions That Can Be Integrated with Factory Equipment
One important aspect of these demonstrations is the effort to integrate AI-supported vision with existing factory infrastructure. It was noted that dedicated controllers for 3D vision could be used, or that these functions could be integrated with industrial robots and programmable logic controllers (PLC), opening the way to reusing some existing equipment instead of building an entirely new production cell.
The demonstrations also included the Cognex In-Sight 3900 system equipped with the Qualcomm Dragonwing platform, an industrial vision system designed to perform high-precision inspection operations directly on the line. Demonstrations of AI-powered vision sensors developed by Mitsubishi Electric and Sony Semiconductor Solutions were also presented, alongside an industrial vision system from Nikon featuring functions for tracking random picking operations.
What Is Changing in Practice?
These developments point to a shift toward moving part of robot setup from manual programming to automated object recognition. This is important for environments in which products change continuously, but it does not eliminate the need to calibrate the system, verify model accuracy, and ensure operational safety. The success of the application will also remain tied to the quality of the training data, lighting conditions, and the integration of the cameras with the robot and controllers.
certi.news analysis: The practical value of these demonstrations lies not merely in adding a camera to a robot, but in attempting to address one of the most complex automation challenges: handling non-identical products in a real production environment. However, the available material does not provide figures on recognition accuracy, picking speed, or long-term operating results, so it is not yet possible to assess the operational difference between these solutions on the basis of standardized performance indicators.