Page 52 - Read Online
P. 52

Page 117                      Bernardo et al. Intell Robot 2021;1(2):116-30  I http://dx.doi.org/10.20517/ir.2021.10

               Robots need to efficiently create semantic models of their environment (semantic maps). One way that has
               proved to have shown great value in representing the information of the environment where robots work, is
               through semantic maps. These combine semantic, topological, and geometric information into a compact
               representation [6,7] . Existing semantic maps need to evolve from task-specific representations to models that
               can be dynamically updated and reused in different tasks. This is one of the major limitations of these ap-
               proaches. Moreover, the ontologies developed to date are not reusable, being one of the major limitations in
               this strategy. Ontologies should move towards a more homogeneous structure and easy interchangeability
                                                          [8]
               between different structures in order to be reusable . Recently, robots are pouring into home environments,
               and thus the need to communicate with humans is increasing. The tasks of robots are not only to navigate in
               an accurate geometrical space but also to understand the indoor environment and share common semantic
               knowledge with people. Consider the task of fetching a cup of coffee. If a robot had only a representation of
               the environment through a metric map, it would have to search in a crude way all over the environment until
               it found the cup. If more knowledge were added through semantics to the robot, such as the probability of
               the cup being in each room, the search could be guided from locations with high probability to locations with
               lower probability. In short, with the evolution of systems and artificial intelligence (AI), ontologies become a
               great solution to make domain knowledge explicit and remove ambiguities, enable machines to reason, and
               facilitate knowledge sharing between machines and humans, focusing on a new generation of intelligent and
               integrated technologies for smart manufacturing.

               Currently, in robotics, the most used middleware is the ROS (https://www.ros.org). This is the standard mid-
               dleware for the development of robotic software, allowing the design of modular and scalable robotic architec-
               tures. There is a framework in ROS called ROSPlan (http://kcl-planning.github.io/ROSPlan/) that provides
               a collection of tools for AI planning, namely ROSPlan. It has a variety of nodes which encapsulate planning,
               problem generation, and plan execution. It possesses a simple interface and links to common ROS packages.
               To date, this framework does not yet have an adequate interface for semantic queries, thus lacking a general
               standardized framework for working with ontologies, natively supporting symbolic logic and advanced rea-
               soning paradigms. In this sense, the paper proposes a framework that integrates a domain specific home
               environment ontology with a task planner (ROSPlan), translating the objectives coming from another agent
               (robot or human) into executable actions by the robotic agent. Two reasoning systems for task planning were
               developed, which are based on ontologies The first system uses the MongoDB (https://www.mongodb.com/)
               database, while the second system uses the domain specific ontology home environment that is proposed in
               this paper.

               The paper is structured as follows. In the next section, the related work is reviewed. The design methodology
               section introduces the reasoning systems presented in this work. In the results section, the structure of the
               ontology is presented, along with the results obtained from the different proposed reasoning systems. In the
               validation and discussion section, the developed ontology is validated and discussed. Finally, the conclusions
               section presents the conclusions of the work and the future way forward.


               2.   RELATED WORK
               Ontologies are a powerful solution for acquiring and sharing common knowledge. Ontologies represent a
               common understanding in a given domain, promoting semantic interoperability among stakeholders, because
                                                                                [3]
               “sharing a common ontology is equivalent to sharing a common world view” . All concepts in an ontology
               must be rigorously specified so that humans and machines can use them unambiguously, empowering robots
               to autonomously perform a wide variety of tasks in a wide variety of environments.

                                                                                       [9]
               Depending on their level of generality, different types of ontologies can be identified ; among many types,
               we can identify as the main ones:
   47   48   49   50   51   52   53   54   55   56   57