Modern robotic picking is moving beyond neat rows and perfect lighting conditions. In this Automate 2025 conversation, Vlad and Dave sit down with Kevin Wu from Siemens to explore how Simatic Robot Pick AI Pro is tackling the messy reality of warehouses and factories. They discuss how the new edge architecture with the Simatic IPC BX 59 A and an NVIDIA GPU lifts pick rates to well over one thousand picks per hour, why multiple suction patterns matter for stability on large or flexible items, how camera agnostic support opens the door to new vision hardware, and why transparent objects are no longer a limitation in many applications.

This episode also dives into digital thread and digital twin workflows using Siemens Process Simulate. These tools allow teams to test new products and layouts virtually before any hardware changes are made, helping reduce commissioning risk and shorten the path to production. The discussion highlights an on-booth demonstration that combines a robot with a secondary camera and a vision language model to identify products and read packaging details such as expiration dates. It is a clear example of how multimodal AI can complement traditional industrial vision systems.

A major theme throughout this conversation is resilience. In real operations, products are rarely placed perfectly. Pallets shift, orientations vary, and lighting changes throughout the day. Traditional rules-based vision systems often struggle when small variances accumulate. Kevin explains how model-free 3D picking localizes unknown objects in clutter, selects stable suction patterns based on measured dimensions, and keeps production moving without forcing operators to maintain perfect alignment.

For manufacturers in consumer packaged goods and medical devices, this is a meaningful advancement. It enables greater product variety and frequent SKU changes while maintaining engineering control. The difference is that the picking logic adapts to what the system sees rather than expecting the environment to remain static.

We also talk about practical evaluation and proof of concept. Siemens runs application testing at its Berkeley, California lab where customers can send sample parts for quick feasibility checks. A short video of their parts being picked can provide the confidence needed to move forward with a pilot project while minimizing cost and risk. For quality inspection and defect detection, Siemens also offers an Inspector station capable of learning from as few as twenty samples to identify defects in real time.

The discussion closes by looking at the future of digital manufacturing. Digital thread tools make it possible to simulate robots from multiple brands, test new configurations, and evaluate throughput virtually. Combined with edge AI and NVIDIA vision language technology, this creates faster experimentation cycles, improved reliability, and measurable gains in uptime and throughput.

Kevin’s key message is clear. Manufacturers do not need to replace existing automation to explore the benefits of AI. Start with one process, validate performance, and build from there.

Timestamps
00:00 Welcome and why real-world picking matters
00:40 Introduction to Pick AI Pro and new throughput capabilities
01:30 Multi suction patterns for stable handling of large items
02:20 Camera agnostic approach and transparent object handling
03:30 Selecting components for high-temperature environments
04:15 Use cases in consumer packaged goods and medical applications
06:45 Digital twin and digital thread with Siemens Process Simulate
08:30 Feasibility testing and customer demos at the Siemens lab
10:30 Vision language model for product identification and labeling
12:10 Evaluating with real parts and rapid testing cycles
14:20 Siemens Inspector for defect detection and visual inspection
15:40 Key takeaways and future outlook

References and Resources Mentioned
Siemens Simatic Robot Pick AI Overview
https://www.siemens.com/global/en/products/automation/topic-areas/tia/future-topics/simatic-robotics-ai.html

Siemens Press Release on Simatic Robot Pick AI Pro
 https://press.siemens.com/global/en/pressrelease/siemens-presents-future-intralogistics-simatic-robot-pick-ai-pro-enables-machine

Siemens Simatic IPC BX 59 A Industrial Edge Device with NVIDIA GPU
 https://www.automationworld.com/products/data/product/55287446/siemens-ag-siemens-simatic-ipc-bx-59a-industrial-edge-device

Siemens IPC BX 59 A Operating Instructions
 https://support.industry.siemens.com/cs/attachments/109972660/ipcbx56a_and_ipcbx59a_operating_instructions_enUS_en-US.pdf

Universal Robots Example with Simatic Robot Pick AI
 https://support.industry.siemens.com/cs/document/109822788/simatic-robot-pick-ai-with-universal-robots-ur5

Zivid Transparent Object Imaging Information
 https://www.zivid.com/zivid-omni-engine-transparency

https://blog.zivid.com/zivid-omni-engine

Siemens Digital Thread Overview and Tecnomatix Process Simulate
 https://www.sw.siemens.com/en-US/digital-thread/

https://plm.sw.siemens.com/en-US/tecnomatix/

NVIDIA Vision Language Model Resources
 https://docs.nvidia.com/nim/vision-language-models/latest/introduction.html

https://developer.nvidia.com/blog/vision-language-model-prompt-engineering-guide-for-image-and-video-understanding/

Hosts
Vlad Romanov is an electrical engineer and manufacturing consultant who leads Joltek and co-hosts the Manufacturing Hub Podcast. He focuses on practical strategies for SCADA, MES, and data-driven operations.
Learn more at https://www.joltek.com
YouTube Channel https://www.youtube.com/channel/UC6JpBeS_6JhUwfGF8RgLCIQ

Dave Griffith is a manufacturing consultant and long-time co-host of Manufacturing Hub. He helps teams align operations, engineering, and leadership around the projects that move the needle in real production environments.

Guest
Kevin Wu from Siemens discusses Robot Pick AI Pro and related digital thread workflows across robotics and vision. Learn more about Siemens automation and software at
https://www.siemens.com

https://www.sw.siemens.com