Learn from Experience: The Manufacturing Way

This week: More productive spaces, Robotic foundation models, 3D-printed titanium, Shale 4.0, recycling tech, 3D reconstructions and Mesh code generation, and AI-powered pneumatic cylinder monitoring

Nov 23, 2025

∙ Paid

Shop Talk

Capturing this week’s zeitgeist

“It’s not just a building. This is a space that is a tool for our employees to be more productive, to be more collaborative, and really help deliver the Ford+ plan,” said Jim Dobleske, Ford Land chair and CEO
“We’re inviting them back into a space that is a tool to help them do their best work. And that best work tends to come with collaboration with other teams,” Dobleske said.

Ford’s new “The Hub” headquarters in Dearborn stands as a physical representation for the rejuvenation of the American industrial heartland, moving decisively away from the rigid, 1950s-era corporate structures to secure a new generation of engineering talent. By replacing traditional offices with collaborative “neighborhoods,” high-end scratch kitchens, and walkable outdoor campuses, Ford is shedding the austere image of legacy manufacturing in favor of a dynamic, amenity-rich environment reminiscent of Silicon Valley. Will it work?

Elsewhere in the heartland: GE Appliances awards more than $150 million in contracts to U.S. suppliers for new laundry manufacturing plant /GE Appliances/

If you found value in this issue, please forward it to a friend, share it on social media, or post it in your organization’s chat.

Assembly Line

This week’s Industry 5.0 breakthroughs and frontier technologies of the built world.

🧠🤖 π*0.6: a VLA that Learns from Experience

Physical Intelligence has introduced 0.6 (Pi Star 0.6), a robotic foundation model that achieves a major leap in reliability by utilizing a new training methodology called Recap (Reinforcement learning with Experience & Corrections via Advantage-conditioned Policies). Moving beyond standard imitation learning, Recap allows the 5-billion parameter Vision-Language-Action (VLA) architecture to “practice” by learning from expert corrections and autonomous reinforcement learning, utilizing a value function to distinguish between successful and failed actions. This approach more than doubled throughput and halved failure rates, enabling the robot to robustly execute complex, long-horizon tasks—such as folding diverse laundry, assembling shipping boxes, and making espresso—continuously for hours in real-world environments.

Read more at Physical Intelligence, arXiv, and X

Physical Intelligence Valued at $5.6 Billion in New Funding /Bloomberg/
It’s worth noting that despite many ideas coming together in the latest 𝜋*₀.₆ release, what it seems to completely avoid is (1) the use of simulation for training and (2) world modeling /Siddharth Ancha on X/ while Meta chief AI scientist Yann LeCun is leaving to create his own startup focused on world models /CNBC/
ACT-1: A Robot Foundation Model Trained on Zero Robot Data /Sunday Robotics/ using glove-based data collection producing higher quality data than teleop on contact-rich tasks /CEO Tony Zhao on X/
Flexion Reflect v0 - Towards Generalizable Robot Autonomy /Flexion/ while raising $50M to Build the Brain of Humanoid Robots /Flexion/
Google-Backed Apptronik in Talks to Raise Funding at $5 Billion Valuation /The Information/ while Google DeepMind Hires Former CTO of Boston Dynamics as the Company Pushes Deeper Into Robotics /Wired/
Jeff Bezos Creates A.I. Start-Up Where He Will Be Co-Chief Executive, called Project Prometheus, focusing on artificial intelligence for the engineering and manufacturing of computers, automobiles and spacecraft. /NYT/ Bezos previously invested in Physical Intelligence.

🖨️⌚ Mapping the future with 3D-printed titanium Apple Watch cases

This year, all Apple Watch Ultra 3 and titanium Apple Watch Series 11 cases are 3D-printed with 100 percent recycled aerospace-grade titanium powder, an achievement not previously considered possible at scale. Every team at Apple rallied behind a shared ambition. The polished mirror finish on Series 11 had to be pristine. Ultra 3 had to maintain its durability and lightweight form to meet the demands of everyday adventurers. They both also had to be better for the planet without compromising performance, and use the same or better-quality materials.

Using the additive process of 3D printing, layer after layer gets printed until an object is as close to the final shape needed as possible. Historically, machining forged parts is subtractive, requiring large portions of material to be shaved off. This shift enables Ultra 3 and titanium cases of Series 11 to use just half the raw material compared to their previous generations.

☁️♨️ Finland’s Big Idea: Turning Data Center Heat Into Power

As AI and data centers consume more power worldwide, Finland is finding a way to turn the problem into a solution. Tom Mackenzie reports from Helsinki, where energy company Helen captures excess heat from Equinix’s data centers and reuses it to warm local homes, cutting emissions and costs at once. With Microsoft and others now expanding similar projects, Finland’s model could redefine what it means to build a sustainable digital economy.

🛢️ Shale Oil’s Next Revolution Should Worry OPEC

✍️ Author: Javier Blas

The next phase of the revolution — call it shale 4.0 — is an engineering arms race to improve the so-called recovery factor. Increasing the ratio even by a single percentage point is a prize worth billions of dollars over the lifetime of thousands of wells in Texas, New Mexico, North Dakota and Colorado. “The best place to find oil is where you already know you’ve got oil,” Chevron Corp. Chief Executive Officer Mike Wirth tells me in an interview in New York. “We know where the oil is. If we left 90% of the oil behind, it would be the first time in history that we didn’t figure out how to do it.”

♻️ The Recycling Breakthrough That Doesn’t Need Sorting

What if every piece of plastic waste, like bottles, bags, even clothes, could be rebuilt from scratch, no sorting required? Not just melted and reshaped, but broken down to pure chemical building blocks and made new again. That’s what a new wave of recycling tech promises. Northwestern University has built a catalyst that can zero in on a single type of plastic in mixed waste and break it down. No sorting, no problem. In South Korea, a 2,000°C hydrogen plasma torch is cracking mixed plastics in milliseconds, turning them into valuable building blocks for brand-new plastics. And in France, an enzymatic recycling plant is transforming previously unrecyclable polyester textiles back into virgin-quality plastic feedstock. We’ve been sold the recycling dream for decades, but the reality? Most plastic still ends up in landfills or the ocean. Could these breakthroughs finally turn recycling from marketing lie into working reality?

MeshCoder: LLM-Powered Structured Mesh Code Generation from Point Clouds

Reconstructing 3D objects into editable programs is pivotal for applications like reverse engineering and shape editing. However, existing methods often rely on limited domain-specific languages (DSLs) and small-scale datasets, restricting their ability to model complex geometries and structures. To address these challenges, we introduce MeshCoder, a novel framework that reconstructs complex 3D objects from point clouds into editable Blender Python scripts. We develop a comprehensive set of expressive Blender Python APIs capable of synthesizing intricate geometries. Leveraging these APIs, we construct a large-scale paired object-code dataset, where the code for each object is decomposed into distinct semantic parts. Subsequently, we train a multimodal large language model (LLM) that translates 3D point cloud into executable Blender Python scripts. Our approach not only achieves superior performance in shape-to-code reconstruction tasks but also facilitates intuitive geometric and topological editing through convenient code modifications. Furthermore, our code-based representation enhances the reasoning capabilities of LLMs in 3D shape understanding tasks. Together, these contributions establish MeshCoder as a powerful and flexible solution for programmatic 3D shape reconstruction and understanding.

New Product Introduction

Highlighting new and innovative facilities, processes, products, and services

🦾🤖🦿 Digit Moves Over 100,000 Totes in Commercial Deployment

For us here at Agility Robotics, we’ve always dedicated ourselves to moving past the novelty of humanoids and focused on providing an essential operational tool and solution for our customers. Today we celebrate a crucial milestone as Digit, our general-purpose humanoid, has moved over 100,000 totes at GXO’s Flowery Branch facility. True industrial validation requires proving a robot’s capacity for high-volume, reliable throughput, effectively establishing a clear, long-term Return on Investment (ROI) and proves that the automation can be scaled successfully across the enterprise. Having a true general-purpose humanoid is the linchpin, but what does that look like and how did we get there?

🧠🎛️ Festo AI-Powered Pneumatic Cylinder Monitoring Cuts Machine Downtime

Festo has introduced its AI-enabled monitoring and predictive maintenance system Festo AX Motion Insights Pneumatic. AX Motion Insights Pneumatic is a cost-effective, easy-to-use solution for preventing costly unscheduled machine downtime through its monitoring of pneumatic cylinders for wear and anomalies. This AI-based industrial app ensures operations are more reliable, efficient, and ultimately more profitable.

The system uses AI to detect anomalies in the behavior of pneumatic drives and cylinders without the need for additional sensors. It provides a “health score” as a “risk of failure” indicator for each cylinder. Connectivity is achieved via a PLC function block. The system is compatible with actuators from different manufacturers and has specific connectivity details for Siemens, Rockwell, and Beckhoff controllers.

🇨🇳 The 27th China Hi-Tech Fair showcases China’s innovation development

Core China’s Strategic Equipment Debuts at the China Hi-Tech Fair /PR Newswire/

TradingView chart — The global market for food-grade CO₂ is dominated by a small group of massive industrial gas corporations (Linde, Air Liquide, Air Products, Messer Group) that control both production and distribution. Why America is Running Out of CO₂ /YouTube/ before the holiday season.

Business Transactions

This week’s top funding events, acquisitions, and partnerships across industrial value chains.

📑 2026 Manufacturing Industry Outlook sees Renewed strategic focus and targeted technology investments could be essential to maintaining a competitive edge in 2026 /Deloitte/

🇺🇸 Archetype AI Raises $35M to Scale Deployment of Physical Agents to Solve Real-World Problems

Archetype AI, the Physical AI company, announced $35 million in Series A funding led by IAG Capital Partners and Hitachi Ventures, with participation from new and existing investors including Bezos Expeditions, Venrock, Amazon Industrial Innovation Fund, Samsung Ventures, Systemiq Capital, E12 Ventures, Higher Life Ventures, and others. Archetype AI is also introducing new tools to build and deploy Physical Agents that sense, understand, and act in real-world environments.

The new funding will enable Archetype to accelerate scaling the Archetype platform, expand the Physical Agent capabilities, and invest further in frontier research and development to advance Newton’s ability to interpret, reason, and act in the physical world. The company is releasing new research results demonstrating the state-of-the-art capabilities of Newton for physical signal-language fusion, which allows the model to generate continuous time series signals from language descriptions. With this research, Archetype is making the next step beyond understanding to acting and manipulating the physical world.

🇺🇸 Lumafield Secures $50 Million Growth Capital Facility from Silicon Valley Bank

Silicon Valley Bank (SVB), a division of First Citizens Bank, announced it has provided a $50 million growth capital facility to Lumafield, a pioneer in advanced manufacturing technology that delivers AI-powered manufacturing intelligence to help companies improve product quality, accelerate development, and automate operations. The company’s industrial computed tomography (CT) scanners and machine learning technology are embedded in production lines at some of the world’s most recognized brands in the medical device, consumer packaged goods, automotive, and battery sectors.

🇫🇷💿 NcodiN Secures €16M Seed Round to Break AI’s Main Bottleneck with the World’s Smallest Laser

NcodiN, the deep-tech startup pioneering optical interposer technology with integrated nanolasers, has secured €16 million in oversubscribed Seed financing round. The funding will propel NcodiN from R&D to industrial scale, enabling to develop product development, key engineering hires to support the industrialization of its technology in a CMOS pilot line on 300 mm wafers, and the build-out of its supply chain and customer partnerships. The equity round was led by MIG Capital, with participation from Maverick Silicon, PhotonVentures, and Verve Ventures, alongside continued support from existing backers Elaia, Earlybird, and OVNI.