The Cutting Edge of Research & Development

Physical AI is no longer an abstract research goal—it is rapidly transitioning from the lab to industry. Over the past two years, three converging advances have unlocked the pace of progress:

1. Vision-Language-Action foundation models — Multimodal models capable of linking text, vision, and control (such as RT-2 and OpenVLA) have demonstrated that general-purpose learning can transform multiple modes into action, enabling robots to both reason and understand intent.

2. Simulation and tactile learning at scale — New tactile models, deformable-material simulators, and reinforcement-learning pipelines—like PhysTwin and DiffSkill—allow robots to acquire dexterous grasping and folding skills from large synthetic datasets rather than years of physical trial-and-error.

3. Academic maturity and early commercial deployment — University research in deformable-object manipulation has reached reproducible baselines, and startups are now adapting these techniques for packaging, recycling, and manufacturing tasks. The field is crossing from theory into applied engineering.

Together, these shifts make this moment uniquely ripe for industrial Physical AI. The data, compute, and models exist; the methods to perceive and predict soft-material behavior now function; and the commercial need—on-demand, sustainable, localized production—is acute. CreateMe sits at that intersection, translating frontier research into deployable manufacturing systems.

At CreateMe, we’re applying this new class of intelligence to one of manufacturing’s hardest problems: deformable-material assembly. Apparel is our proving ground because it demands the precision of robotics, the nuance of touch, and the speed of mass production—across more than ten billion possible permutations of fabric type, garment and silhouette, material stretch, and construction method. No other category combines this degree of physical variability with such massive global scale.

To tackle the challenge of automating soft-material assembly at industrial scale, we began by rethinking how materials join. We built Pixel™, a bonding process engineered for precision, repeatability, and automation, replacing many of the mechanical constraints of sewing with a digital workflow that drives speed, consistency, and sustainability. That foundation set the stage for the next step: teaching machines to execute the process autonomously at production scale.

By combining Physical AI with this bonded-assembly process and our modular MeRA™ robotics platform, we’re proving that machines can not only handle fabrics—they can form them, bond them, and adapt in real time.

The leap from mechanical sewing to intelligent bonding marks a fundamental turning point in soft-goods manufacturing; a paradigm shift on par with the transformer breakthrough in language models. It transforms how clothing is made and unlocks new ways to automate complex, deformable materials.

Most attempts at automating apparel begin by imitating needle-and-thread. We started somewhere else. CreateMe’s architecture treats garments as 2D and 3D forms to be shaped and bonded, not pierced and stitched. That choice matters.

• Single-sided access — By designing assembly steps that operate from one side, we eliminate interference and occlusion that make multi-sided sewing difficult for robots. Single-sided access simplifies perception and control.
• 3D molds and tooling — Purpose-built forms hold geometry so fabric behaves predictably. Static fixturing converts a chaotic, deformable substrate into a controlled state where learning thrives.
• Grippers — Fabric is chaotic; our grippers tame it. We use end-effectors that offer granular control over porous and stretchable materials, providing the physical dexterity needed to execute complex AI policies.
• Adhesive joining — Bonding removes the mechanical constraints of needles and enables programmable, functional joining patterns. Adhesives let us encode stretch, comfort, and durability directly into the bond design.

This is more than a manufacturing preference—it’s the right substrate for Physical AI. By re-engineering the problem around single-sided access, 3D forms, dexterous grippers, and bonding, we reduce the dimensionality of the control problem, creating a more structured environment where manipulation policies can reliably complete the task.

In redefining the problem, we redefined the possibility: a world where intelligent robotics can shape materials with the same ease that software shapes information.

Our first generation combined specialized tooling with machine vision to locate parts and guide precise placement. It worked; it shipped. Physical AI takes the next step: instead of encoding every motion, we teach generalizable skills for soft-material handling—flattening, aligning, folding, loading, attaching—through multimodal data, teleoperation, and simulation.

The act of learning becomes the new form of manufacturing—each iteration a conversation between data and matter.

The system learns to adapt to wrinkles, stretch, drift, and variation in real time; the “how” is no longer brittle code but an evolving skill. Once a capability is learned, it travels. A pocket-placement skill developed in T-shirt assembly becomes the basis for accurate gusset placement in intimates; a sleeve-loading policy accelerates waistband alignment. The library compounds.

Soft goods introduce partial observability, contact-rich manipulation, and non-linear material response—precisely the complexity that Physical AI must master. That’s why apparel is the ultimate proving ground.

If a machine can perceive, predict, and control cloth—especially while forming to 3D molds and bonding on single-sided access—it develops a transferable grammar for deformables. What we learn here becomes the foundation for how machines will handle any flexible material in the future.

There’s also a systems reason. Physical AI transforms the architecture of manufacturing itself: distributed, near-consumer, and on-demand. When learning turns assembly into software, small high-throughput lines can be placed near demand and updated over the air. Lead times collapse; overproduction falls; sustainability becomes a property of the system.

Mastering fabric manipulation is a significant step toward understanding the physics of physical interaction, which is fundamental to embodied intelligence.

Bonded assembly isn’t just a different join—it’s a new way to encode function. Micro-patterned adhesives allow us to specify stretch and recovery, drape, and feel at the bond itself. Physical AI closes the loop: perception informs placement, force and vision feedback inform cure and finish, and predictive models optimize bond topology to match the unique mechanics of every material.

Each bond we create is both a joint and a data point—a record of how matter and intention meet.

This turns motion into knowledge. Each grasp, fold, and alignment expands a living library of skills and outcomes; each outcome refines the next grasp. Over time, deformable-object assembly becomes editable—more like software than tooling.

While apparel offers the perfect proving ground for bonded automation and Physical AI, its implications extend far beyond fashion. The same principles of precision material handling, adhesion, and adaptive assembly apply to technical textiles across multiple industries.

CreateMe’s technologies are already adaptable to automotive interiors (bonded seat structures, soft panels, and composites), medical devices (biocompatible textile assemblies and wearables), furniture and upholstery (3D-contoured soft-surface bonding), and aerospace and defense (advanced laminates and structural fabrics).

Each of these categories presents the same challenges—variable material properties, complex 3D geometry, and tight-tolerance requirements. The combination of bonded-assembly architecture, robotic dexterity, and learning-based control positions CreateMe’s platform as a foundational technology for any industry working with soft matter.

This is not just the future of clothing—it’s the future of how we build with flexible materials.

Physical AI doesn’t just make us better; it makes us faster. Every new garment category creates new data; every new dataset strengthens shared skills; every shared skill reduces time-to-capability for the next category.

That flywheel compresses development cycles, expands the range of supported fabrics and constructions, and multiplies the commercial readiness of each new line. Learning reduces capital intensity; software scales what hardware began.

The result is an accelerating loop between research and production—a living system that grows smarter with every unit built.

CreateMe’s Physical AI roadmap is not theoretical—it’s underway. Over the course of 2026, we will transition from foundational research and simulation to direct integration within our automated apparel assembly lines.

In the first half of 2026, we will be focusing on establishing the core infrastructure for learning-based manipulation. In late 2026, these systems will intersect with our production-grade robotic assembly lines, automating high dexterity taste in initial product categories—starting with T-shirts and intimates.

The result will be a fully integrated, software-driven, end-to-end manufacturing line—one capable of learning and improving autonomously through data.

This integration marks the practical realization of CreateMe’s vision: a commercial Physical AI system operating inside a live factory environment, accelerating us toward full-scale adaptive manufacturing.

CreateMe’s research channel will share credible updates on deformable manipulation, bonded assembly, and learning-based automation, along with insights from our work in teleoperation, simulation, and sensing—advancing the broader field while building toward scalable commercialization.

The near term is clear: advance bonded assembly with learned manipulation; broaden fabric and product coverage; ship adaptive robotic lines that improve with every run. The long view: factories that learn—small, fast, updatable, and close to the customer.

We see a decade ahead where every product—from clothes to car seats to medical textiles—will be assembled by systems that adapt and learn, rather than follow static rules.

CreateMe exists to make that future real.