3D AI picks random parts

Robots usually fail at the factory job humans take for granted: grabbing one usable part from a messy pile of identical-looking pieces. Siemens is now showing a 3D vision system doing that job on a live production line in Chengdu. (siemens.com) The problem is called bin picking: a robot has to find one object in an unsorted heap, figure out its position in space, and choose a place to grip it without hitting the bin or other parts. Siemens says its system uses a 3D camera, artificial intelligence software, and a vacuum gripper to calculate pick points for arbitrarily positioned objects. (siemens.com) Siemens introduced its latest commercial version, Simatic Robot Pick AI Pro, on March 6, 2025, at LogiMAT in Nuremberg. The company said the software is pre-trained, works without a computer-aided design model of each object, and returns six-degree-of-freedom gripping poses in milliseconds. (press.siemens.com) That matters because random-pile picking has been one of the hardest gaps in warehouse and factory automation. Siemens says the software is aimed at single-piece order picking and other jobs where parts arrive in mixed orientations and manual picking still dominates. (press.siemens.com) Siemens ties the product to a broader push into what it calls Industrial Artificial Intelligence: software that uses shop-floor data to adjust production and logistics in real time. On its Industrial Artificial Intelligence page, the company says these tools are meant to support adaptive manufacturing, planning, and maintenance across the value chain. (siemens.com) The Chengdu setting is not incidental. Siemens’ Industrial Automation Products Chengdu production and research base was named a World Economic Forum “Sustainability Lighthouse” in December 2023, and Siemens says the site uses digital twin and industrial artificial intelligence tools in day-to-day operations. (w1.siemens.com.cn) Siemens has also been expanding that China footprint. In June 2023, the company said it would invest 140 million euros, about 154 million dollars at the time, to expand its Chengdu digital factory and add 400 jobs. (english.news.cn) The technical pitch is that the robot does not need a perfect catalog of every item before it starts. Siemens says the software was trained on synthetic and real-world data, then refined for cluttered scenes where standard object-recognition models often make mistakes. (siemens.com) Siemens’ own application examples show where it expects the system to land first: intralogistics, warehousing, e-commerce, and online grocery operations that still rely on workers to pick high volumes of items. In those setups, Siemens says staff shift from repetitive grabbing to operating stations, monitoring cells, and troubleshooting. (support.industry.siemens.com) So the Chengdu demo is less about a robot arm moving one part than about a factory test for a stubborn automation problem. If the system keeps working outside trade-show booths, the next measure will be how many manual picking stations it can replace or reassign. (press.siemens.com)