Simulating deformable objects under complex interactions remains one of the biggest challenges in robotics, especially when transferring real-world behavior to virtual environments for training manipulation algorithms. Traditional simulators rely on predefined physical models or purely data-driven dynamics, but rarely couple robot actions, environmental effects and material properties in an integrated manner. This limits accuracy, stability and generalization. In this context, SoMA, a neural simulator based on 3D Gaussian Splats, has emerged that promises to revolutionize soft-body manipulation by unifying deformable dynamics, environmental forces and robot joint actions in a single latent neural space, enabling end-to-end simulation from real to sim.
SoMA, which stands for Soft-body Manipulation Simulator, represents a significant advance over previous approaches. Instead of relying on explicit physical equations, the simulator learns to model interactions over 3D Gaussian Splat representations, allowing stable and generalizable long-horizon manipulation, even beyond observed training trajectories. This translates to a 20% improvement in resimulation accuracy and generalization on real-world robotic manipulation tasks, such as long-horizon cloth folding. The key is that SoMA not only simulates object behavior, but also conditions dynamics on robot control signals, creating a closed loop between perception, action and environment.
From a technical perspective, SoMA employs a neural latent space where both geometric and dynamic properties of the deformable object are encoded. Gaussian Splats, originally popular in 3D rendering, are used here as representation units that can evolve over time under the influence of external forces and robot commands. This approach eliminates the need for simulation models with manually tuned physical parameters, which often fail to generalize to new configurations or unforeseen forces. Additionally, because it is a model trained on real data, SoMA captures complex phenomena such as friction, nonlinear elasticity and folds, which are difficult to model analytically.
The relevance of SoMA extends beyond the lab. In the industrial domain, the ability to accurately simulate manipulation of soft materials —from textiles to food or medical components— opens the door to automating processes that currently require human intervention. Logistics, manufacturing and service robotics companies can benefit from simulators that allow training virtual agents before deploying them in real environments, reducing costs and risks. However, developing and integrating neural simulation solutions like SoMA requires a combination of expertise in artificial intelligence, custom software development and a robust cloud infrastructure. This is where Q2BSTUDIO positions itself as a strategic partner.
Q2BSTUDIO is a software and technology development company offering specialized services in custom applications, artificial intelligence, cybersecurity, cloud AWS/Azure, Business Intelligence (Power BI) and AI agents. For a company looking to adopt capabilities similar to SoMA, the process starts with developing a tailored simulation model, which can be built on cloud platforms like AWS or Azure to scale training of complex neural networks. Q2BSTUDIO's experience in custom software ensures that the solution fits exactly the specific workflows and requirements of each business, whether in robotics, logistics or any other sector.
Moreover, integrating artificial intelligence into industrial processes demands strong cybersecurity measures to protect sensitive data and trained models. Q2BSTUDIO provides cybersecurity services including pentesting, audits and cloud infrastructure protection, ensuring that neural simulation solutions are safe from internal and external threats. On the other hand, using Business Intelligence via Power BI allows monitoring simulator performance, analyzing accuracy metrics and optimizing computational resources in real time. AI agents, one of Q2BSTUDIO's most innovative areas, can be integrated to automate decision-making based on simulation results, creating continuous improvement loops.
The impact of SoMA on soft robotics is undeniable, but its potential only materializes if companies have the right technological ecosystem. The combination of advanced neural simulators with scalable cloud services, data analytics and cybersecurity forms the foundation of the next generation of intelligent automation. Q2BSTUDIO, with its portfolio ranging from custom application development to AI agent implementation, is ready to accompany organizations on this journey, whether by improving existing simulations or creating entirely new solutions from scratch.
In summary, SoMA represents a qualitative leap in soft-body simulation, and its neural approach offers a path towards more accurate and generalizable robotic manipulation. To leverage these innovations, companies need technology partners that understand both the complexity of AI models and the demands of production environments. Q2BSTUDIO, with its expertise in artificial intelligence, cloud AWS/Azure, cybersecurity and BI, becomes the ideal ally to turn ideas like SoMA into real applications that generate value.
Neural simulation is not just a research trend; it is a practical tool that, when properly implemented, can drastically reduce development costs and increase the reliability of robotic systems. Whether in cloth folding, flexible part assembly or manipulation of biological materials, SoMA marks the beginning of a new era where simulation and reality converge thanks to the power of AI and custom software. Q2BSTUDIO offers precisely that: the ability to build bridges between the virtual and the real, with personalized solutions that drive business innovation.
