In this paper we report on the design and implementation of a metamodel for the Unified Robot Description Format (URDF) on the MM-AR metamodeling platform. The goal of this approach is to provide the foundation for the integration of digital twins of robots as specified in URDF with conceptual enterprise models. In particular, the MM-AR platform and its underlying meta2 model permit the inherent integration of principles from spatial computing for enabling the representation and manipulation of conceptual models in spatial environments. Ultimately, this can be used to link behavioral specifications of robots on various abstraction levels with the structure and physical properties of robots as given by URDF.
Robotics is a complex technical domain combining multiple disciplines, such as mechanical engineering,
computer science and specific application domains. On a general level, we can distinguish in robotics
between the structure of robots and the behavior that they exhibit. The structure of robots can today be
well described thanks to the Unified Robot Description Format (URDF), which is supported by various
design tools such as SolidWorks [
As explored in [
Therefore, we propose a combination of URDF with spatial conceptual modeling. The goal is to enable easier model-based design of robotic behaviors, while at the same time allowing users to view or edit the robot state using URDF concepts. To achieve this, we first need to represent URDF using spatial conceptual modeling concepts. For this purpose we rely on a meta2 model for spatial conceptual modeling and a corresponding software implementation in the form of the MM-AR metamodeling platform. This allows us to provide a first demonstration of the interactions between behavioral models and 3D robotic architectures in a pick-and-place scenario.
With this article, we investigate the following research questions (RQs): • RQ1: What added value does a URDF metamodel provide over using URDF files directly? • RQ2: Which URDF concepts and relationships are necessary and suficient for the URDF metamodel to faithfully represent fixed industrial robots and support digital twins? • RQ3: How can spatial conceptual modeling be embedded in the URDF metamodel to align robot structure with process models?
RQ1 evaluates the benefits and trade-ofs of a metamodel compared to a static, file-based approach. RQ2 delineates the core elements and associations the metamodel must capture. RQ3 focuses on integrating spatial conceptual modeling so that structural views are correctly represented in the 3D space.
The remainder of the paper is structured as follows: Section 2 introduces foundations required for describing the approach and related work. Section 3 presents how the URDF metamodel was derived and how it has been implemented in the MM-AR web tool. Section 4 then explores how the approach was applied to the pick-and-place scenario. Finally, in Section 5, we discuss the approach and how it relates to the field of Digital Twins. In Section 6 we conclude the findings and propose further research directions.
In this section we will present the necessary foundations and related work in robotics and URDF, knowledge-based digital twins, as well as spatial conceptual modeling.
Robots come in various shapes and sizes. From small mobile units to robot vacuum cleaners [
For this purpose, the Unified Robot Description Format (URDF) was developed to standardize the
representation of robots and their properties [
URDF itself also covers mobile platforms (e.g., bases with planar/floating joints and wheel assemblies). A complete mobile-robot digital twin additionally needs environment/localization frames (map/odom) and motion constraints, which are, for now, outside of the scope of this work.
URDF further allows representation of the physical aspects of a robot, but not its behavior. There is a gap between low-level definitions of robots and high level tasks. Robot programming itself is a complex undertaking that requires deep knowledge of the field. The use of conceptual modeling in this field aims to abstract from the technical aspects of a specific domain to make it more accessible for domain specialists to define robotic behavior.
According to Singh et al. [
The digital twin concept is widely used in robotics [19]. For the scope of this paper, we introduce the notion of knowledge-based digital twins, in which the models of a digital twin are grounded in a common meta2 model and can therefore be integrated with other conceptual and enterprise models. This integration enables the re-use of knowledge from other model types in digital-twin scenarios. For example, a user may define a pick-and-place task in some enterprise modeling scenarios for a 6-DoF robotic arm. Then, the knowledge-based digital twin can exploit properties of the geometric model to verify that the specified poses are physically reachable; the physical model can assess whether joint eforts remain within predefined constraints; the behavior model can analyze the path required to achieve the goal; and, finally, the rule model can check that the pick-and-place action does not violate system rules, such as entering a safety area. We therefore leverage the capabilities of URDF to analyze and simulate robotic tasks. By representing URDF models using spatial conceptual modeling, we can reference and re-use their concepts and visualizations in other conceptual models such as business process models or other types of enterprise models.
Spatial conceptual modeling integrates principles from spatial computing in conceptual modeling [
Generating URDF files has been extensively studied, with most contributions focusing on the transformation pipeline between modeling tools and URDF artefacts [23, 24]. Schneider et al. present a DSL-based development process for specifying grasping problems, reusing established standards, such as URDF, to define robotic structure and dynamics [ 25]. While this strategy facilitates the integration of emerging standards, it lacks a structured modeling approach to systematically exploit URDF concepts.
Ringe et al. propose a metamodeling approach to classify robot morphologies by introducing a dedicated metamodeling language [26]. Thereby, they can represent a broad spectrum of robot morphologies, i.e., the physical structure, shape, and configuration of robots. However, despite its breadth, this approach does not benefit from the extensive ecosystem of tools that support URDF. MetaMorph therefore requires an explicit mapping back to the URDF format and does not solve the challenge of URDF generation.
With the approach we propose in the following we therefore aim to combine the strengths of the URDF ecosystem with the flexibility of conceptual enterprise modeling. Thus, we want to bridge the gap between robotic morphology definition and conceptual modeling, e.g., for defining robotic behavior.
For integrating URDF with the approach of spatial conceptual modeling we use the meta2 model of the MM-AR approach and the corresponding metamodeling platform. In this way, we will show how URDF can be realized as a meta model using these foundations. We will first briefly present the MM-AR meta2 model and then advance to the derivation of the URDF meta model including the expected user interface.
The MM-AR meta2 model and the according implementation allow a flexible realization of modeling languages based on the approach of spatial conceptual modeling. The main concepts of the meta2 model of MM-AR that will be required for describing the URDF meta model are as follows: • SceneType denotes a template for a container of Classes with some common purpose, similar to diagram types in 2D modeling environments. • Class denotes a template for objects that have a common structure in the form of Attributes. • Attribute denotes a template for a property, which can be attached to SceneTypes or Classes and whose values are constrained through an AttributeTable or an AttributeType. • AttributeTable denotes a template for a tabular structure of properties in the form of two or more
Attributes. • AttributeType denotes a template for constraining the value of an Attribute, e.g., a number, a mass, or a reference to another object. For values other than references to objects, regular expressions are used to constrain the range of values.
For integrating fundamental properties of spatial conceptual modeling, all Classes have pre-defined attributes for locations in diferent coordinate systems, transform attributes, as well as an attribute for the three-dimensional graphical representation. As a consequence, every object that is instantiated based on these concepts can be inherently represented in a 3D modeling environment. However, as will be shown in the next section for robotics, further attributes for spatial information may be needed depending on the application domain.
URDF describes a robot as a set of links connected by joints. These links and joints have associated properties. In our URDF metamodel, links and joints are modeled as Classes. Both have attributes that capture the corresponding URDF properties.
Every concept in URDF has its counterpart in the URDF metamodel, which builds on the MM-AR meta2 model. For example, Origin is an AttributeTable with the Attributes , , , , ℎ, . Using a table serves a dual purpose: it aggregates recurrent concepts and enables the inclusion of a temporal perspective in the model with rows representing diferent states over time.
Figure 1 depicts the URDF metamodel of the link concept. The Link Class has three optional AttributeTables: Inertial, Collision, and Visual. The URDF name property is mapped to the default Name Attribute available for every object in the MM-AR meta2 model; the same mapping applies to joint names.
The Inertial AttributeTable comprises two further AttributeTables: Origin and Inertia and one scalar attribute, Mass. Inertia represents the link’s inertia tensor, Origin specifies the pose of the inertia frame, and Mass records the link mass.
The Visual Attribute table comprises three attribute tables: Origin, Geometry, and Material. Origin is as defined above. Geometry defines the shape of the visual object and can be a sphere, cylinder, or box, or a reference to an external 3D mesh (e.g., STL or glTF). Material may be a color defined by an RGBA value or a reference to an external specification (e.g., a .material file). Some 3D mesh files include embedded materials and textures; in such cases, a separate material attribute is unnecessary. The Collision AttributeTable contains the Geometry and Origin tables.
Figure 2 represents the URDF metamodel of the joint concept in URDF. This is a Class meaning that it is a visual instance. It has seven AttributeTables: Origin, Axis, Calibration, Dynamics, Limit, Mimic and Safety Controller. Parent and Child are references to links. The joint type attribute defines the type of joint that links two parts. This type can be one of the following: revolute, continuous, prismatic, fixed, lfoating, planar. Joint units are derived from the type of selected joint outside of the metamodel. For example, revolute joints uses radians whereas prismatic joint have units in meters. The axis attribute table reuses the x,y,z concept of the origin attribute table. The axis defines the joint axis in the joint frame. The x,y,z components represent a vector and must be normalized. The Dynamics AttributeTable defines the Damping and Friction attributes. They define the physical properties of the joint, that can be leveraged for simulation. The units are derived from joint types. For example, the damping in newton-seconds per meter for prismatic joints and in newton-meter-seconds per radian for revolute joints and the friction in newtons for prismatic joints, in newton-meters for revolute joints. The Limit attribute table defines the Lower, Upper, Efort and Velocity Attribute of a joint. This AttributeTable defines the limits of the joint. The Calibration attribute type defines the Rising and Falling attributes. They specify the joint position at which a calibration sensor or encoder index transitions high (“rising”) or low (“falling”) during homing. The Safety controller defines soft lower limit, soft upper limit, k position and k velocity. The soft limits are not physical limits but limits that can be defined by the user. K position scales the maximum permitted speed based on how close the joint is to the soft limits, while K velocity sets an overall speed ceiling, ensuring smooth deceleration near limits and preventing excessive velocities throughout the motion range. The Mimic attribute table contains the reference to a joint to mimic, the multiplier and the ofset. - Link : Class [1..n] - Joint : RelationClass [0..n]
Link - Name : String [1..1] - Intertial : TableAttribute [0..1] - Collision : TableAttribute [0..1] - Visual : TableAttribute [0..1] - Name : String [1..1] - Type : Attribute [1..1] - Origin : AttributeType [0..1] - Parent : Attribute [1..1] - Child : Attribute [1..1] - Axis : TableAttribute [0..1] - Calibration : TableAttribute [0..1] - Dynamics : TableAttribute [0..1] - Limit : TableAttribute [0..1] - Mimic : TableAttribute [0..1] - Safety control er : TableAttribute [0..1]
The URDF metamodel has been implemented on the MM-AR platform. Although all data structures including the access via a REST-API could be realized, the front-end of the platform still misses some UI components required for editing instances of the URDF metamodel. Figure 3 therefore shows a mock-up of the Visual attribute of a link as the currently used user interface does not yet support the concept of a table of a table as detailed in Section 3.2. The Origin can have multiple rows representing multiple states (i.e., positions) of the visual model.
When loading URDF files into the MM-AR platform, the process as depicted in Figure 4 is initiated. The URDF file is first imported into the MM-AR web client tool and converted to an instance of a SceneType with the URDF metamodel. In this way, the concepts of the URDF file can be directly accessed from other scene instances on the platform. The 3D view can be an AR scene or a VR environment. Users then interact with the controller models that allow or restrict the interaction with the 3D model.
In this section, we illustrate the concepts introduced above through a pick-and-place scenario. Pickand-place is among the most common tasks in robotics [27] and can be generalized across a variety of contexts. In our example, a box is picked from an initial location and placed, after a prescribed rotation, at a target location.
Figure 6 shows the URDF model obtained by converting the original URDF file so that it matches the URDF metamodel defined in the previous sections. The model represents the 6-DoF robotic arm Dobot Magician E62. It consists of a base and six links connected by six joints. The BPMN model in Figure 7 captures the pick-and-place process and is linked to the URDF model in Figure 6. The selected activity can be visualized in augmented reality (AR), as illustrated in Figure 5. The 3D view can be aligned spatially to the robot’s physical position, yielding a ghost-like digital overlay of the real robot.
The URDF schema could be successfully implemented on the MM-AR metamodeling platform. Although the current user interface implementation of the MM-AR web client does not yet fully support all necessary UI components - e.g., the above-mentioned tables-in-tables representations, we can assess that the MM-AR meta2 model provides all concepts required for the implementation of URDF. Further, although MM-AR natively supports the three-dimensional representation of objects - currently in GLTF format only - and URDF is format-agnostic with regard to 3D formats, the sample robot files used so far
A BPMN implementation of the pick-andFigure 7: place using the URDF model in the MM
AR platform were only available in STL format. Therefore, the next step will be to either conduct conversions into GLTF format if possible, or to extend the implementation of MM-AR to also support the STL format.
Robotic tasks inherently unfold in 3D environments, which are often dificult to capture faithfully with traditional conceptual modeling notations.
In our approach, the 3D environment serves as a first-class representation of the digital twin’s current state. Conventional conceptual modeling approaches typically constrain digital twin representations to two-dimensional diagrams or isolated 3D visualizations with limited semantics. By integrating a URDF model, we enhance the ability to represent digital twins in 3D while enriching them with explicit conceptual semantics, e.g. for specifying behavior in relation to the URDF components.
URDF benefits from a mature ecosystem of algorithms and simulation tools. Our metamodel implementation enables these capabilities to be integrated into the MM-AR platform, thereby supporting simulation and analysis scenarios. For example, inverse kinematics can be applied to assess reachability between two poses, and collision checking can verify that no interferences occur for a given task model. These extensions benefit both communities: robotics, which contributes established analyses and simulations, and conceptual enterprise modeling, which provides rigorous domain-specific languages and tooling. In doing so, the approach helps bridge the gap between conceptual modeling and knowledge-based digital twins.
The current metamodel does not yet include proposed URDF extensions such as sensors or end efectors. Nevertheless, its structure permits the addition of non-robotic artifacts and attributes without compromising interoperability with other models or the integration mechanisms.
In this paper, we proposed a URDF metamodel that integrates with other modeling languages based on the MM-AR meta2 model. The metamodel is directly mapped from the URDF XML specification, enabling a faithful representation of robotic architectures composed of links and joints. Each element can be associated with properties such as visual and geometry descriptions, mass and inertia, and joint limits. Mapping URDF concepts to a metamodel allows the instantiation of models representing diverse robotic systems; these instances can then be referenced by other (meta)models to support broader modeling and analysis scenarios. By adopting the URDF metamodel, we reduce the efort required to represent and edit 3D robotic tasks and leverage the existing ecosystem of URDF-based algorithms and analyses. Designing and analyzing robotic tasks digitally could not only accelerate prototyping and development but may also reduce costs compared to testing on physical machinery. We can answer the research questions as follows. (RQ1) The metamodel provides first-class integration with enterprise and process models via typed cross-model references. The inherent spatial anchoring (locations, transforms, and 3D representations) can be consumed in AR/VR. The validation and analysis hooks (e.g., unit constraints, joint-limit consistency, pose reachability) are defined at the modeling level, as well as the queryability and versioning of structure and state beyond file-level XML. And crucially, it maintains interoperability with the URDF ecosystem, while enabling conceptual links to behavior models that raw URDF files do not support. (RQ2) For fixed industrial robots and digital-twin support, the minimal set realized in our metamodel is: • Link with Inertial (Origin, Inertia, Mass), Visual (Origin, Geometry, Material), and Collision (Origin,
Geometry). • Joint (a relation between Links) with type (revolute, continuous, prismatic, fixed, floating, planar), Origin, Axis, Limit, Dynamics, Calibration, Safety Controller, and Mimic, plus a scene-level root link and coordinate frames.
These cover kinematic topology, core dynamics, geometry/meshes, and operational bounds, which we found suficient to import an industrial 6-DoF arm, render it in 3D, attach time-varying poses, and run typical analyses (e.g., reachability/collision) used in digital twins. This implementation aligns with the elements defined by the URDF XML specification. Elements like SRDF groupings or sensors are useful extensions but not required for faithful fixed-robot representation.
(RQ3) We embed spatial conceptual modeling by leveraging MM-AR’s native spatial attributes so that every URDF element has a pose and 3D representation. We store pose rows in AttributeTables (e.g., Origins) so the same robot structure can express diferent runtime states over time along a process. We created typed relations between process activities (e.g., BPMN tasks) and URDF elements (links, joints, target poses). This allows process models to constrain or derive spatial states (paths, keep-out zones, approach poses) and to drive AR overlays, thereby aligning behavioral steps with the robot’s structural and physical semantics.
As next steps, we will implement a working prototype that combines the URDF (meta)model with a process model to specify and execute robotic tasks in a 3D environment. We plan to include concepts from SRDF and Xacro to have an even more complete robotic modeling metamodel.
Financial support is gratefully acknowledged by the SmartLiving Lab (https://www.smartlivinglab.ch/ en/) funded by the University of Fribourg, EPFL, and HEIA-FR.