New Simulation Model Bridges Rigid and Soft Robotics
In the rapidly evolving field of robotics, a persistent challenge has been the gap between rigid and soft robotic systems—two domains that have traditionally operated on separate technological tracks. Rigid robots, with their metal frames and precise mechanics, dominate industrial automation and structured environments. Soft robots, composed of flexible, deformable materials, excel in unstructured, dynamic settings where adaptability and safety are paramount. However, combining the strengths of both—precision from rigid structures and compliance from soft actuators—has been hindered by a critical bottleneck: the lack of a unified simulation framework capable of modeling hybrid systems accurately. Now, a research team from the Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, has introduced a breakthrough solution: a novel coupled modeling approach that seamlessly integrates rigid and soft robotics within a single simulation environment.
The innovation, developed by Zeyang Xia, Jun Chen, Yangzhou Gan, and Jing Xiong, addresses a long-standing limitation in robotic simulation platforms. While tools like Gazebo and Webots have matured for rigid-body dynamics, and SOFA or FEBIO have advanced the modeling of soft, deformable materials, none could natively simulate systems where both types coexist. This disconnect has forced researchers to either oversimplify hybrid designs or rely on fragmented, ad-hoc simulation workflows, slowing down innovation in areas such as medical robotics, human-robot collaboration, and adaptive manipulation.
The team’s approach, published in the journal Robot, introduces a “rigid-soft coupled model” that functions as a middleware layer between existing simulation engines. Instead of building a new platform from scratch, the researchers leveraged the strengths of established tools—Gazebo for rigid dynamics and SOFA for soft tissue simulation—by creating a bridge that allows them to exchange physical data in real time. This strategy not only accelerates development but also ensures compatibility with widely used robotic frameworks, particularly the Robot Operating System (ROS), which serves as the backbone of modern robotics research and deployment.
At the core of the model is a clever abstraction: soft components are represented as a series of rigid segments connected by flexible joints. These joints are not physical parts but computational constructs that capture the elastic behavior of soft materials using principles from finite element modeling. When external forces act on the rigid segments, their resulting motion is fed into the soft simulation engine, which computes the corresponding deformation and reaction forces. These forces are then fed back into the rigid simulation, closing the loop and enabling a realistic, bidirectional interaction between the two domains.
This hybrid representation strikes a balance between computational efficiency and physical fidelity. Full finite element simulations of soft bodies, while accurate, are often too slow for real-time control or interactive applications. By discretizing soft structures into a smaller number of rigid elements, the model reduces complexity without sacrificing essential behaviors such as bending, twisting, and compression. The result is a simulation that runs fast enough for real-time applications while still capturing the nuanced compliance that defines soft robotics.
To validate their approach, the team conducted a series of experiments focusing on a common hybrid configuration: a rigid robotic arm equipped with a soft pneumatic gripper. In one test, they simulated the bending response of a single soft actuator under varying air pressure, comparing the results directly with physical prototypes. The simulation closely matched the real-world behavior across a wide pressure range, from negative (suction) to positive (inflation) pressures. While minor discrepancies were observed—particularly under negative pressure, where physical constraints like reinforcing fibers played a role not fully modeled in simulation—the overall agreement was strong, demonstrating the model’s predictive capability.
The real power of the framework emerged in a more complex scenario: multi-object grasping in a simulated industrial environment. Here, a six-degree-of-freedom Staubli TX90 robotic arm, fitted with a three-finger soft pneumatic hand, was tasked with picking up a variety of irregularly shaped objects transported on a conveyor belt. The simulation environment included realistic physics, friction, and collision dynamics. The results showed that the soft gripper could successfully adapt its shape to conform to different objects, achieving stable grasps without prior knowledge of object geometry—a hallmark of soft robotic manipulation.
This experiment was not merely a demonstration of dexterity; it was a proof of concept for the entire simulation pipeline. It showed that the coupled model could handle the dynamic interaction between a high-speed rigid manipulator and a compliant end-effector, including the transmission of forces, the propagation of deformation, and the feedback necessary for control. The ability to simulate such interactions in a virtual environment before deploying on real hardware significantly reduces development time and risk, enabling faster iteration and more robust system design.
The implications of this work extend far beyond the laboratory. In healthcare, for instance, hybrid robots are being explored for minimally invasive surgery, where a rigid arm provides reach and stability, while a soft end-effector interacts safely with delicate tissues. In manufacturing, soft grippers on rigid arms can handle fragile or irregular products—such as food items or electronic components—without damage. In search and rescue, hybrid systems could navigate debris with rigid mobility bases while using soft appendages to probe and manipulate in confined spaces.
Until now, the design of such systems has been constrained by the inability to simulate their full behavior. Engineers could model the rigid arm’s motion or the soft gripper’s deformation in isolation, but not how they influence each other during operation. The coupled model removes this barrier, allowing for holistic system-level optimization. Designers can now explore trade-offs between stiffness, compliance, actuation strategy, and control algorithms in a virtual sandbox, leading to more intelligent and capable robots.
Moreover, the model’s integration with ROS ensures broad accessibility. ROS has become the de facto standard in both academic and industrial robotics, with a vast ecosystem of tools, libraries, and community support. By building on this foundation, the team has ensured that their work can be easily adopted, extended, and shared. Researchers can plug their own rigid robot models into Gazebo, their soft actuators into SOFA, and use the coupled framework to connect them—without needing to rewrite core physics engines or develop custom communication protocols.
The choice of SOFA as the soft simulation backend is particularly strategic. Originally developed for medical simulation, SOFA excels at modeling complex biological tissues and soft materials under dynamic loads. Its ability to handle large deformations, nonlinear material properties, and contact mechanics makes it ideal for soft robotics. By linking SOFA’s advanced physics with Gazebo’s robust rigid-body dynamics, the team has created a simulation environment that mirrors the complexity of real-world hybrid systems.
Another key advantage of the approach is its modularity. The coupling layer is designed to be agnostic to the specific simulation engines used. While the current implementation uses Gazebo and SOFA, the underlying concept could be adapted to other platforms. This future-proof design ensures that the model can evolve alongside the broader robotics software ecosystem, incorporating new tools and capabilities as they emerge.
The research also highlights a shift in how robotic systems are being conceptualized. Rather than viewing robots as purely rigid or purely soft, the field is moving toward a spectrum of mechanical compliance, where different parts of a robot can have tailored material properties. This biomimetic approach—inspired by animals that combine rigid skeletons with soft muscles and skin—promises a new generation of robots that are more versatile, safer, and more capable of operating in human-centered environments.
From a software engineering perspective, the work represents a mature approach to system integration. Instead of reinventing the wheel, the team embraced existing, well-tested tools and focused their innovation on the interface between them. This philosophy aligns with modern software development practices, where interoperability and reuse are valued over monolithic, proprietary solutions. It also reflects a growing trend in robotics toward open, collaborative development, where advances are shared and built upon by a global community.
The validation experiments further underscore the practical relevance of the research. By comparing simulation results with physical prototypes, the team demonstrated not just theoretical correctness but real-world applicability. The slight deviations observed—such as those due to unmodeled constraint fibers—are not flaws in the model but opportunities for refinement. They provide clear pathways for future work, such as incorporating more detailed material models or adaptive meshing techniques to better capture thin-walled structures.
Looking ahead, the coupled model opens the door to more sophisticated applications. For example, it could be used to train machine learning models for hybrid robot control, where reinforcement learning agents learn to manipulate objects in simulation before being deployed on real hardware. It could also support the design of variable stiffness actuators, where the compliance of a joint can be actively changed during operation—a capability that blurs the line between rigid and soft even further.
In addition, the framework could facilitate the development of digital twins for robotic systems—virtual replicas that mirror the physical state of a robot in real time. Such twins are increasingly important in industrial automation, where predictive maintenance, performance optimization, and remote monitoring are critical. With the ability to simulate both rigid and soft components accurately, a digital twin could provide unprecedented insights into a robot’s behavior, wear, and performance.
The work also has educational value. By providing a unified simulation environment, it lowers the barrier for students and new researchers entering the field of hybrid robotics. They can experiment with complex systems without needing deep expertise in both rigid dynamics and soft material modeling. This democratization of access can accelerate innovation and broaden participation in robotics research.
In summary, the rigid-soft coupled model developed by Xia, Chen, Gan, and Xiong represents a significant step forward in robotic simulation. It bridges a critical gap between two previously isolated domains, enabling the design and testing of hybrid robots with unprecedented fidelity and efficiency. By leveraging existing tools and integrating them within the widely adopted ROS framework, the team has created a solution that is not only technically sound but also practical and accessible. As the demand for adaptive, safe, and intelligent robots grows, this work provides a foundational tool for the next generation of robotic systems.
Zeyang Xia, Jun Chen, Yangzhou Gan, Jing Xiong, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Robot, 10.13973/j.cnki.robot.200050