NVIDIA AI Podcast, Ep. 257

Title: How Siemens Is Bringing AI to Factory Floors
Host: Noah Kravitz (NVIDIA)
Guest: Matthias Loeskil, Head of Virtual Control and Industrial AI, Siemens Factory Automation
Date: May 20, 2025

Episode Overview

This episode spotlights Siemens' groundbreaking efforts, in partnership with NVIDIA, to integrate AI across manufacturing environments worldwide. Matthias Loeskil discusses the origins and goals of this collaboration, the real-world challenges and opportunities in bringing AI to factory floors, and concrete examples of transformative AI-powered solutions. He also provides insights into future directions, particularly in robotics, generative AI, and the increasing importance of on-premise AI for manufacturing.

Key Discussion Points and Insights

1. Siemens and NVIDIA: The Partnership & Vision

• Collaboration Beginnings:
  • Partnership started in 2022, focusing on the “industrial metaverse” by linking Siemens Xcelerator with NVIDIA Omniverse for high-fidelity digital twins and real-time simulation ([03:10]).
  • "Essentially, this is about high fidelity digital twins and real-time simulation tools." — Matthias ([03:13]).
• Shared Vision:
  • Bringing AI to manufacturing to accelerate digital transformation:
    "We want to make AI more relevant for manufacturing sectors, making it more applicable to accelerate digital transformation." — Matthias ([03:37])
• Synergies:
  • NVIDIA’s strengths in accelerated computing and AI, Siemens’ dominance in industrial automation — leading to numerous industrial-grade AI deployments.

2. Why the Push for AI in Manufacturing Now?

• Technological Readiness:
  • Recent advancements (deep learning, generative AI) make the tech more adaptable for industry ([04:40]).
• Demand Drivers:
  • Labor shortages, retiring skilled workers, supply chain disruptions, and reshoring to high-cost countries intensify the need for automation ([05:19]).
  • "Factories and production facilities are moved back to high cost countries, which... means you have to increase the level of automation as well." — Matthias ([05:21])
• Objective:
  • Employ the latest in AI and digital twin technology for a new level of intelligence and efficiency.

3. Siemens Xcelerator Explained

• Multi-faceted Platform:
  • A software platform, an ecosystem for partner services, and Siemens’ commitment to guide customers through digital transformation ([06:40]).

4. AI's Role: Humans and Automation Side-by-Side

• AI Replace vs. Augment:
  • AI and robots take on repetitive, dangerous, or lower-value tasks; humans focus on thoughtful, higher-value work ([07:54]).
• Collaboration:
  • Increasingly collaborative setups (cobots and eventually more sophisticated robots working alongside humans).

5. Key Challenges in Industrial AI

• Trust and Adoption Barriers:
  • Trust Issues: 40% of manufacturers see AI as "not trustworthy" ([09:05])
    • “We're not talking about recommendation of your favorite movie... We are talking about health and safety of human workers. So you better get it right.” — Matthias ([09:33])
  • Talent Shortage: 92% report a lack of AI-skilled experts ([10:39])
  • Scaling Failure: Only 16% of companies achieve their AI goals; 70–80% of industrial AI projects don’t deliver expected ROI ([10:39], [12:22])
• Main Reasons:
  • AI projects often remain at proof of concept; operationalization and scaling are major hurdles.
  • Need for standardized software infrastructure.

6. Real-World Use Cases

a. Inspecto: Democratizing AI-Driven Quality Inspection

• Purpose:
  • Out-of-the-box, AI-powered quality inspection adaptable even for companies without AI expertise ([12:29]).
• Key Features:
  • Pre-configured hardware/software, pre-trained models on millions of industrial images.
  • Fast onboarding: 20 "good" sample images suffice, setup in under an hour ([15:36]).
  • Not limited to specific industries — works best for rigid objects like metal, plastics, electronics ([16:16]).
• Customer Example:
  • mtcon, a small German company, automated inspection of delicate electrical connectors without needing data science expertise ([17:07]).
    • “They can adapt it and change the inspection settings on their own... Even the operators are able to use the system.” — Matthias ([17:54])
• NVIDIA Partnership:
  • Inspecto runs on Siemens’ industrial PCs with NVIDIA GPUs for high-speed, low-latency inference ([19:11]).

b. Audi: Scaling AI for Automotive Manufacturing

• Use Case:
  • Automated detection and removal of weld spatter in Audi car body manufacturing.
  • 5,000 weld spots per car, over a million per day per factory — sample checks formerly manual, costly, and labor-intensive ([19:55])
• Solution:
  • Audi’s data science team trained models to automate inspection and removal, then worked with Siemens to scale for real-world deployment.
  • Siemens provides an industrial AI suite for standardized deployment and management, leveraging NVIDIA hardware and Triton inference server ([21:18], [24:00])
    • “We were able to achieve up to 25-fold acceleration in AI execution directly on shop floor.” — Matthias ([24:41])

7. Making AI More Accessible: Pre-Trained and Foundational Models

• Adoption Approach:
  • Industrial AI benefits from pre-trained/foundational models but still often requires domain-specific fine-tuning ([25:47]).
  • Generative AI (e.g., large language models) will also require adaptation to specific industrial contexts ([26:51]).

8. Future Roadmap for Siemens and NVIDIA Collaboration

a. AI-Enhanced Robotics

• Piece-Picking Robots:
  • "Simatic Robot Pick AI Pro" solution for robots to grasp unpredictable objects — critical for logistics/warehousing ([27:59]).
  • Recent work focuses on using synthetic data and NVIDIA Isaac Sim/Ominverse for simulation-to-reality gap bridging ([29:36]).

b. Generative AI and On-Premise Copilots

• Industrial Copilots:
  • GenAI-powered assistants (copilots) that help engineers, operators, and technicians with tasks from code generation to maintenance troubleshooting ([31:31]).
  • Deployed locally (on-premise) due to security and data privacy requirements, powered by NVIDIA NIM microservices ([34:48]).
  • "It's like a super experienced digital colleague you could imagine, available around the clock..." — Matthias ([31:37])

c. Emerging Challenge: Agentic AI

• Towards Autonomous Agents:
  • Next frontier is agentic AI — not just assistants, but agents who execute and reason independently.
  • Industrial settings require agent AI to have safety guardrails and deterministic behavior, which remains an active research challenge ([34:37]).

Notable Quotes & Memorable Moments

• “We want to make AI more relevant for manufacturing sectors, making it more applicable to accelerate digital transformation.” — Matthias ([03:37])
• “Factories…are moved back to high cost countries, which…means you have to increase the level of automation as well, boosting efficiency.” — Matthias ([05:21])
• “We're not talking about recommendation of your favorite movie. We are talking about critical setups, big investments into machinery. We are talking about health and safety of human workers. So you better get it right.” — Matthias ([09:33])
• “Only 16% of companies achieve their AI related goals … 70 to 80% at least of AI projects in industry … fail to deliver the return of investors that was actually expected.” — Matthias ([10:39],[12:22])
• “Inspector comes with pre-trained AI models inside … you just present a few good samples in the end and you don’t even have to show pictures of defects.” — Matthias ([15:18])
• “We were able to achieve up to 25-fold acceleration in AI execution directly on shop floor, close to where the critical processes occur.” — Matthias ([24:41])
• “It's the same thing in the office space nowadays. Right. So absolutely not like you must use it, but you get more used to it and you get more efficient and maybe you save half an hour a day.” — Matthias ([32:36])
• “How do you give them [AI agents] certain guardrails that come close to something like a deterministic behavior which is often needed in industrial settings.” — Matthias ([34:37])

Key Timestamps

• 03:10 — How and why the Siemens-NVIDIA partnership began
• 04:40 — Why the manufacturing sector is ready for AI now
• 06:40 — Explanation of Siemens Xcelerator
• 07:54 — Vision of humans & robots working together
• 09:05 — Major challenges (trust, talent, operationalization)
• 12:29 — Introducing Inspecto: democratizing AI-driven quality control
• 15:36 — How Inspecto adapts to different industries and products
• 17:07 — mtcon: Small business case for low/no-code AI adoption
• 19:34 — Audi use case: AI-driven inspection at scale
• 21:18 — Why standardized infrastructure matters in scaling industrial AI
• 24:00 — Achieving 25x acceleration for Audi use case with NVIDIA tech
• 25:47 — Foundational/pretrained models in industrial contexts
• 27:18 — Near-term roadmap: robotics, generative AI
• 29:36 — Using synthetic data & simulation-to-reality gap
• 31:31 — Siemens Copilots: on-premises generative AI assistants
• 32:36 — User adoption: operator feedback on copilots
• 34:37 — Agentic AI: future promise and challenges
• 34:48 — Running copilots with NVIDIA NIM microservices on-premises

Conclusion & Further References

The episode presents a compelling look into how Siemens, in close collaboration with NVIDIA, is pioneering real, scalable, and accessible AI deployment in manufacturing, tackling challenges from technical trust to workforce skills. Concrete products like Inspecto and large-scale rollouts with customers like Audi demonstrate how the theoretical potential of AI is rapidly becoming operational reality on the shop floor. Listeners are encouraged to search “Siemens Industrial AI” or “Siemens Xcelerator” for more information as Siemens continues to expand its offerings, especially in generative AI and robotics.