‍ALLSIDES launches the 3D Digital Twin Factory to deliver SimReady 3D data for Physical AI at industrial scale

Built on NVIDIA accelerated computing, the automated pipeline transforms real objects into physically grounded 3D Digital Twins that robots can train on in NVIDIA Isaac Sim and any other simulators, with early adoption by Mistral AI, NVIDIA, and Industrial Next, and already in use at TUM and Heidelberg University.

BERLIN, Germany, October 20, 2026 — NVIDIA GTC Berlin — ALLSIDES, the company building the 3D data infrastructure for Physical AI and World Models, launched the ALLSIDES 3D Digital Twin Factory: an automated, industrial-scale production system that transforms physical objects into SimReady 3D Digital Twins for training robots and Physical AI systems in simulation. Each Digital Twin captures an object's geometry with sub-millimeter accuracy, its measured PBR materials and its physical properties, such as mass, colliders and friction coefficients, so that what a robot learns in simulation holds up in the real world.

Closing the Real2Sim2Real gap

Before a robot picks up a cup, grasps a pan or assembles a server rack, it has practiced the task millions of times in simulation. That training is only as good as the virtual world it happens in. If a digital object is shaped slightly wrong, or its surface behaves differently from the real one, the skill may not transfer to physical deployment. Engineers call this the Sim2Real gap.

Closing it takes 3D data that is physically accurate, diverse and available in large volumes. Until now, developers have had to choose: synthetic datasets offer scale but not physical realism, while studio-captured datasets offer quality but not diversity or scale. The ALLSIDES 3D Digital Twin Factory is designed to remove that trade-off.

From physical object to a SimReady, physically grounded Digital Twin in minutes

An object, whether a sneaker, a power tool, a piece of fruit, an industrial part or a package, is placed in the ALLSIDES capture rig. Minutes later it exists as a physically grounded 3D Digital Twin. ALLSIDES has automated the full chain of capture, reconstruction, quality control and delivery, so Digital Twins are produced continuously rather than one at a time by specialists. All assets are SimReady and can be plugged directly into NVIDIA Isaac Sim and other simulators of choice.

The pipeline runs on NVIDIA accelerated computing. During capture, NVIDIA RTX A1000 GPUs handle real-time AI inference and multi-view image decoding. Reconstruction runs on NVIDIA RTX PRO 5000 and RTX PRO 6000 Blackwell workstations, where ALLSIDES' data orchestration and neural reconstruction software uses NVIDIA CUDA, NCCL, OptiX, NVML and TensorRT.

Every Digital Twin is SimReady and ships with:

+ Sub-millimeter geometric accuracy across hundreds of object categories

+ Fully relightable, measured PBR materials

+ Physical properties, including measured mass and colliders, and derived friction coefficients

+ High-resolution, sub-millimeter-accurate SDF colliders (for dexterous tasks)

+ High-performance convex decomposition colliders to train robots at massive scale

+ Agentically derived, physically accurate center of mass, inertia tensor and restitution

+ Structured metadata per object: category, dimensions and material properties

+ Multi-view consistent captures, optimized for training multimodal and 3D-native AI models

+ Export formats including USDZ, GLB and OBJ, ready for downstream ML pipelines

Ready for NVIDIA Omniverse and Isaac Sim

ALLSIDES Digital Twins are designed to be directly consumable within NVIDIA's Physical AI ecosystem. As SimReady assets with accurate geometry and PBR materials, they function as high-fidelity scene assets created using NVIDIA Omniverse libraries for physically accurate 3D simulation and OpenUSD-based workflows. Within NVIDIA Isaac Sim, an open-source robotics simulation framework and a core Omniverse library, ALLSIDES assets can be placed into robot training environments, so AI models and robot policies learn from physically accurate representations of real-world objects instead of synthetic approximations.

Validated by Meta and Stanford

In 2025, Meta Reality Labs Research partnered with ALLSIDES, then operating as Covision Media, to power the capture pipeline behind the Digital Twin Catalog (DTC), the “world’s highest quality dataset for object reconstruction research”. In the peer-reviewed DTC paper by Meta and Stanford, ALLSIDES-captured data was benchmarked against leading datasets including ShapeNet, Objaverse, Stanford-ORB and OmniObject3D, and was the only data to combine real-world capture, multi-view consistency, full PBR materials, high geometric accuracy and relightability. The DTC is used today by companies and research teams training multimodal AI models and robots.

ALLSIDES, through its own computer vision, ML and AI research and development, went beyond that first study and made its 3D reconstruction and data pipeline SimReady.

Scaling production

The factory is built on the same pipeline ALLSIDES has used to produce hundreds of thousands of 3D assets for brands including adidas, Nike, Zara and Amazon. It has set out to deliver millions of 3D Digital Twins across several hundred different object categories, with new data released monthly and flexible licensing.  

“Every major advance in AI has been powered by better data, and Physical AI is no different,” said Franz Tschimben, CEO and co-founder of ALLSIDES. “Robots learn in simulation, so simulation has to be true to the real world. The 3D Digital Twin Factory delivers SimReady Digital Twins of real objects at industrial scale.”

“Accuracy in simulation starts with accuracy in capture,” said Burkhard Guessefeld, CTO and co-founder of ALLSIDES. “Running our reconstruction on NVIDIA accelerated computing lets us measure geometry, materials and physical properties for every object, and deliver them in formats robotics teams can drop straight into Isaac Sim.”

Early partners and adopters of SimReady data include NVIDIA, Mistral AI, Franka Robotics, Industrial Next, the Technical University of Munich and Heidelberg University.

Early adopters praise quality and scale of the SimReady data library and platform

“ALLSIDES provides the gold standard of SimReady and physically grounded 3D data in the world today. The fact that ALLSIDES provides this at scale thanks to its 3D Digital Twin Factory and 3D Data Platform is a big step forward for the entire Physical AI and robotics industry.”
– Dr. Tomas Hodan, AI and Robotics Scientist at Mistral AI

“Physical AI in retail and logistics depends on robots that understand how real objects behave — how a sneaker compresses under grip, how a garment folds, how a package is best handled at speed. Simulation is only as useful as its fidelity to the real world, and that fidelity starts with the data. ALLSIDES has built the pipeline that produces physically accurate, SimReady digital twins at industrial scale, running on NVIDIA accelerated computing and dropping directly into NVIDIA Isaac Sim — from the same capture system already trusted by adidas, Nike and Zara for commerce. That convergence of commerce-grade 3D data and robot training data from a single pipeline is exactly where physical AI meets retail at scale.”
– Azita Martin, VP and GM of AI for Retail, CPG and QSR at NVIDIA

“SimReady data is extremely critical for robotics, the most foundational asset that will close the sim2real gap and make robots useful in the real world. The ALLSIDES 3D Digital Twin Factory is a big step for researchers and companies working to solve the sim2real gap.”
– Dr. Gilwoo Lee, CTO of Industrial Next

“People used to say data was the new gold. As we enter the era of spatial intelligence, 3D data and 3D assets are becoming the new gold as they provide an accurate and faithful reproduction of what our 3D world actually looks like. And this is where ALLSIDES excels in terms of scalable and faithful reconstruction of thousands and thousands of objects.”
– Prof. Daniel Cremers, Professor at the Technical University of Munich, Germany, and Chair of Computer Vision & Artificial Intelligence.

“Ultra-high-fidelity, SimReady assets are critical to closing the sim-to-real gap in robotics. ALLSIDES not only delivers these assets, but its end-to-end system generates digital twins within minutes, enabling the first truly scalable 3D digital twin platform. This technology will be key to stress-testing edge cases in simulation and training the next generation of generative AI models.”
– Prof. Carsten Rother, Professor at Heidelberg University in Germany, heading the Computer Vision and Learning Lab.

About ALLSIDES

ALLSIDES transforms physical objects into physically grounded 3D Digital Twins through automated 3D acquisition and neural reconstruction. Its 3D data powers digital commerce and content for global brands, and its SimReady 3D data trains Physical AI and robotics systems, closing the Real2Sim2Real gap. Founded in 2024 by Franz Tschimben, Burkhard Guessefeld and Harald Oberrauch, ALLSIDES is headquartered in Brixen (Bressanone), Italy, with offices in New York and San Francisco. Its team of 50 represents 18 nationalities. More at allsides.tech.

Media contact: Sarah Prugger, ALLSIDES · sarah.prugger@allsides.tech · marketing.hq@allsides.tech. Images, video and interviews available on request.

Resources

• ALLSIDES 3D Digital Twin Factory: allsides.tech/dataplatform
• Meta Digital Twin Catalog: dtc.projectaria.com
• DTC research paper (Meta and Stanford): arxiv.org/abs/2504.08541

NVIDIA, the NVIDIA logo, CUDA, NVIDIA Isaac, NVIDIA Omniverse, OptiX, RTX and TensorRT are trademarks and/or registered trademarks of NVIDIA Corporation in the U.S. and other countries. All other trademarks are the property of their respective owners.