Autonomous robots are used in a variety of tasks and context; however, there are still issues with their performance and reliability in real-world scenarios. This research focuses on the use of declarative knowledge to improve the dependability of robots in complex, dynamic environments. We propose to leverage Model-Based Systems Engineering through ontologies so that systems can flexibly respond to unexpected situations while maintaining performance standards. Moreover, to enhance reusability and granularity, we rely on formal conceptualizations based on Category Theory. This integration of knowledge representation and mathematical models seeks to aid decision-making to adapt robot behavior in the presence of contingencies.
CEUR ceur-ws.org
Autonomous robots have revolutionized various domains by performing complex tasks with endless possibilities. However, there are still open issues in dependability and trust. To ensure successful deployment in unstructured, real-world scenarios, robots require a better understanding of their environment and the tools to act and react properly, even in the presence of high uncertainty.
This research focuses on using declarative knowledge to represent and reason about the
cross-cutting elements of the robot, such as its mission, its design, and its operation in open
environments. By exploiting this information at runtime, robots can adapt their structure and
replan actions to keep pursuing their goals, even in unexpected situations. Ontologies are used
at runtime to ensure mission fulfillment within expected performance, thereby increasing robot
lfexibility, explainability, and eficacy.
and Automation [
The combination of knowledge representation and reasoning with abstract mathematical system models based on Category Theory (CT) can significantly enhance the performance, dependability, and trustworthiness of autonomous robots. Using ontologies at runtime allows the robot to reason about the system, its environment, and its behavior, so it has a better comprehension of the situation it faces. This helps the autonomous robot to make more informed decisions. The categorical theoretical model provides a formal foundation to ensure consistency along the changes that the knowledge base, or the system itself, can sufer during robot operation. This can aid in the accuracy and reliability of the system as it strives for the system to behave as expected.
Robotics has shown enormous possibilities in a variety of tasks and environments. However, there are still open issues that compromise the dependability of autonomous robots. Our motivation stems from the challenges of bridging the reality gap between robots used in controlled environments and their deployment in unstructured, real-world scenarios. To enable successful deployment in such scenarios, it is essential to equip robots with the necessary tools to act and react properly to undesired events with high uncertainty. To overcome these challenges, we shall endow the robot with a better understanding about what is happening, what capabilities the robot has and what tools the robot can use to reach its goals. Our research studies the hypothetical benefits of (i) formalizing system models -including both software and hardware- and (ii) exploiting such models at runtime. These models can potentially improve robot awareness, increasing explainability and resiliency towards unexpected contingencies.
We aim to use robust models at runtime as a tool to deliver a better understanding of the situation and react properly in unstructured environments. This can be condensed as the research question: “Is it possible to enhance robot’s understanding—about itself, its mission and its environment—from a systemic perspective?”
As our approach is motivated by system models and mathematical abstract formalisms, the following question arises: “Does system models increase robot dependability, i.e., its ability to fulfill user needs, in complex missions and/or in presence of contingencies?”
Lastly, as we envision knowledge representation and reasoning as a way to leverage explicit engineering knowledge during robot operation, we question its impact on robot performance: “How ontological reasoning about complex system models afect robot decision-making?”
The aforementioned research questions can be translated into specific objectives to create a software asset that can be used at runtime and is based on a CT-grounded abstract formalism: • Analyze existing approaches for system modeling and knowledge-driven robots. • Explore the relation between self-awareness and models. • Formally define constituents of robot operation and its relationships in an abstract mathematical model, e.g., mission, design, behavior, perception, actuation, environment, etc. • Develop a layered ontology based on robot formal models. • Build reusable software assets that exploit robot ontologies at runtime. • Demonstrate the validity of the approach at least in three robotic deployments, including simulated and real robots.
Given the multidisciplinary approach in this work and the desire to obtain generality without losing sight of the engineering application, we follow a mixed methodology between the scientific method and the engineering process. First, we follow a preliminary analysis phase to characterize the problem, identify existing approaches, and identify key issues, research areas, and technologies of potential relevance. This study allows us to establish the design principles for Formal System Model. This model shall be refined into several applications to prove its applicability to diferent robot contexts and its reusability. We shall use testbeds to validate the research conducted in both simulated and real robot deployments.
As this research focuses on several fields, there is no common solid mainstream ground to depart from. To guarantee a steady progress, we follow the snowflake methodology , in which we start with a simplistic approach to the analysis-development-validation process and iterate over it to add complexity and generality. Some important milestones in our approach upon which we iterate are (i) the formalization of robot concepts in an abstract mathematical framework, (ii) the grounding of models in ontologies to exploit them at runtime, and (iii) the evaluation of how runtime reasoning impacts robot performance, fault tolerance, and explainability.
This section aims to provide a clear and concise summary of the research conducted. In a broad sense, it establishes a landscape on how ontologies can be formalized by category theory to provide reusable software assets to increase robot dependability.
We conducted a systematic review of ontology-enabled processes for trustworthy robot
autonomy [
Following the ontological approach, we extended the Teleological and Ontological Model for
Autonomous Systems (TOMASys) framework, a meta-model developed to provide concepts
for modeling the functional knowledge of autonomous systems. This work aims to reuse
system models on a variety of deployments. In particular, we applied this approach to an
underwater robot [
To overcome these issues, we propose the use of Category Theory (CT), a framework to
support system modeling and behavioral analysis for complex Systems of Systems (SoS). CT is a
general theory of mathematical structures. It was invented in the 1940s to unify and synthesize
diferent areas in mathematics [
In conclusion, we believe that explicit formal knowledge can support autonomous robot operation in unstructured environments and the necessary engineering processes. There are still many open issues with respect to the reliability, safety, and explainability of meeting the expectations of researchers and industry. However, the steps taken towards enhancing robot autonomy using ontologies have proved how promising this approach is. Category theory provides a neutral, universal framework to formally depict structure and behavior; to represent and reason about complex systems with many interconnected components in an abstract sense. Such formalization can increase ontological soundness and reusability, taking a further step towards convergence and harmonization.
This work was partially supported by the ROBOMINERS project with funding from the European Union’s Horizon 2020 Research and Innovation Programme (Grant Agreement No. 820971), by the CORESENSE project with funding from the European Union’s Horizon Europe Research and Innovation Programme (Grant Agreement No. 101070254), and by a grant from Programa Propio of Universidad Politécnica de Madrid.