Social robotics is a research area that aims to develop models that allow robots and humans to interact naturally. One of the factors that compromises the evolution of social robotics is the difficulty in integrating cognitive and robotic systems, mainly due to the volume and complexity of the information existing in a dynamic environment. In this work, we are proposing an ontology, named OntPercept, for formalize the communication between the cognitive and robotic systems. Thanks to this ontology is possible to model the environment perception using the information captured by the sensors present in the robotic system. It will be shown that OntPercept simplifies the development, reproduction and comparison of experiments associated with social robotics.
Robots are capable of millimeter-precision movements by performing repetitive tasks operating in structured environments where objects are in known and predictable locations. Thus, it is not surprising that robots are used more in manufacturing operations, such as painting and welding, instead of operations where, diversity of actions, direct contact with humans and changes in the environment are part of the system requirements.
Despite its complexity, the demand for the use of robotic agents in environments
other than manufacturing is a necessity of modern society. The World Robotics 2017
report [IFR 2017] indicates a cumulative volume of around 27 billion U.S. dollars in the
service robots sales forecast for the 2018-2020 biennium. This upward trend in the use of
service robots requires a growth in research involving Human-Robot Interaction (HRI),
in particular in the sub-area defined by
This work contributes in this process by developing an environment directed to
social robotics involving the areas of ontology, HRI and cognitive science. For this,
in our previous work, an architecture, named “Cognitive Model Development
Environment (CMDE)” was proposed [
Copyright © 2020 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
An important component of CMDE is the OntPercept ontology, that is responsible for formalizing the communication of the environment perception data from Cognitive to Robotic System. Since its first version in 2017 1, two were the focus of evolution: optimization of objects access (bypassing performance issues of the producer-consumer information exchange model used with the triple store) and inclusion of the memory concept as a resource for robot long term interaction.
The organization of this article follows the steps used to create an ontology. The hierarchy, concepts and properties of the OntPercept are detailed in Section 2. In Section 3 is presented the test environment. In Section 4, the conclusion of this work is presented.
The ontology presented in this article emphasizes the need to formalize the information that transits between the processing nodes defined by CMDE architecture (Fig. 1). The idea is to enable the cognitive system to receive perception information from the environment with a high level of cognition.
The development of an ontology is accelerated by the use of appropriate modeling
methodologies and tools. This work uses the Ontology Development 101 methodology
[
1Initially the ontology was called OntSense, the name was changed to ontPercept to distance itself from the sensor notion and reflect the essence of this ontology: environment perception.
An important feature of the OntPercept is adherence to the standards of international
organizations with the aim of minimizing effort and providing a better dissemination of the
results. In Fig. 2, the relationship with top-level ontologies is presented as well as the
main concepts of the new ontology. The upper block of this figure presents the top
ontologies, namely: SUMO [
In the lower block, the key concepts of OntPercept are presented. The concept
OntPercept:RobotPerceptionEvent represents the super class used as a basis for defining
the environment perception. All other concepts derive or maintain a relationship with this
super class. Next, we will explore the OntPercept:RobotMemory concept. Details of the
other concepts can be obtained at [
OntPercept:RobotMemory concept: Although not directly associated with environment
perception, the OntPercept:RobotMemory2 concept represents the initial step in the
implementation of robotic long-term interaction systems. In this work, information retrieval
is associated with the use of retrieval cues3 stored together with the information
content. The idea is to use association mechanisms similar to those used in the memorization
process of humans [
2In this work, the term “memory” represents the structures and processes associated with the storage and retrieval of information or experiences.
3A retrieval cue is a hint or clue that can help retrieval.
In Table 2, it can be found the properties of the OntPercept:RobotMemory concept. Basically, the core is represented by a relationship with an instance of the SUMO:Entity class, which is super class of SUMO:Object and SUMO:Process. The other relationships represent cues that allow the retrieval of the memory register: capture time, event duration and the sensory environment perception at capture time.
An important memory concept feature is associated with the forgetting process.
In humans the action of forgetting information from the Long Term Memory (LTM) can
be explained using the following theories: interference, retrieval failure and lack of
consolidation [
Unfortunately, robot memory is finite, and judicious use of memory is advisable. Two theories mentioned above may assist in the memory management process. The Interference Theory can be used to replace one robot’s experience with another similar, where the robot has succeeded in its operation. On the other hand, the absence of pointers to the memory register can be used as a catalyst for garbage collector operations in the records, similarly to Retrieval Failure Theory. Note that by using ontology to formalize memory operations, axioms can be created to identify these obsolete records. Subsequently, reasoning tools4 identify these records and triggers actions for their removal.
In Fig. 4, the relationships established between OntPercept classes and top-level
ontologies are presented. Classes from SUMO are flagged in green and
The final step established by the “Ontology Development 101” methodology involves the generation, storage and retrieval of object instances. The strategy adopted to enable the handling of instances is presented in Fig. 5, with the definition of three processing nodes: Cognition Node, responsible for the implementation of the cognitive system, SPARQL Server Node, which implements a database with sensory information and Robotic Agent Node, which directly controls all sensors and actuators of the robotic agent.
Basically, the robotic agent obtains information from the environment and routes it to the cognitive system using a predefined communication protocol such as REST/HTTP. The cognitive system (“Cognition Node”) processes the environment perception information, makes decisions and sends actions to the robotic agent via socket communication.
In order to reduce end-user effort, it was created two APIs to access the “SPARQL Server Node”. These APIs provide a simpler interface for using the technologies mentioned here (see Fig. 5). The APIs implement insertion and retrieval operations of perception events captured by the robotic agent.
The use of a triple store for sending the environment perception represents a bottleneck in this deployment. In one cycle, dozens of information are stored in the triple store by the Robotic Node and then, they are read and immediately removed by the Cognition Node. In order to optimize this process only information associated with long-term memory will be kept in the triple store, the other triples will be serialized and sent directly to the Cognition Node.
The test is accomplished by the exercise of the ontology in a controlled environment adhering to CMDE architecture. More specifically, given a usage scenario, the OntPercept must offers resources to represent the exchange of perception information between the cognitive and robotic systems.
Robot House Simulator (RHS) [
The steps to create an experiment involve: defining the desired behavior for the robotic agent, defining Soar production rules implementing the behavior, and finally, exercising the results in RHS simulator. The execution of a experiment can be visualized in the video available at https://youtu.be/rPqfQvReXDo.
This article proposed an ontology to establish the communication between cognitive and robotic systems. This ontology was incorporated into CMDE architecture and a simulator for social robotics, called “Robot House Simulator” (RHS). All these elements are used together to design, develop and validate systems geared towards social robotics. All configuration items generated in this development are released into GitHub repository (https://github.com/helioaz/OntPercept) with the GNU GPL v3.0 license.
The results achieved represent a important step in the development of an integrated
environment for the development of cognitive systems with application in social robotics.
This framework is constantly evolving and, as such, a catalyst for future proposals, e.g.:
inclusion of new elements in OntPercept; interface with others cognitive architectures
[
This research is partially funded by the CAPES/MEC, the CNPq, the FAPESP under the grant 2017/01687-0 and the INCT-INSAC (National Institute of Science and Technology) under the grant CNPq 465755/2014-3, FAPESP 2014/50851-0 and FAPESP 2018/25782-5.
IFR
(2017). Why service robots are booming worldwide. ifr.org/ifr-press-releases/news/ why-service-robots-are-booming-worldwide. [Online; accessed 30 jun. 2020].
Kotseruba, I., Gonzalez, O. J. A., and Tsotsos, J. K. (2018). A review of 40 years of cognitive architecture research: Focus on perception, attention, learning and applications.
Computing Research Repository - CoRR, abs/1610.08602v3.
Laird, J. E. (2012). The Soar Cognitive Architecture. AISB Quarterly, 171(134):224–235.