We present here one of the demonstrators we implemented as part of our research on the integration of robots in smart cities, where an autonomous mobile platform is employed for monitoring and assessing the Health&Safety rules in a smart o ce environment in combination with a centralised infrastructure for data integration, processing and reasoning. Monitoring of the status of infrastructures and environments still presents a number of challenges in terms of knowledge acquisition and representation, as data need to constantly be re-evaluated due to their high dynamism. Common solutions, ranging from human monitoring to sensor deployment, fail in exibility, costs and, in the case of large scale scenarios, scalability. We focus on the idea that autonomous mobile agents can be used as moving sensors deployed by a larger, knowledgebased infrastructure, where the central unit collects and reasons over the information produced by the agents. In particular, the paper presents HanS, the Health&Safety inspector, with the goal of showing that applications integrating robots as data consumers and collectors can be deployed thanks to a combination of state-of-the-art semantic and robotics technologies.
Monitoring the status of infrastructures and environments is a task common to many domains including, for instance, the smart cities scenarios. In these cases, a centralised infrastructure is in charge of collecting and integrating data from a variety of sources (energy consumption, transport and mobility data, citizens opinions etc.) to support decision making and resource optimisation [?]. In our research, we are looking into integrating autonomous mobile agents into this \system of systems", with the idea that they can provide valuable services in urban scenarios, but also that their capabilities can be improved by using the vast external knowledge provided within the city infrastructure.
Managing data in these scenarios is still a challenging task in terms of knowledge acquisition, as data can change and/or lose validity in time, resulting in the central infrastructure to become outdated relatively quickly, hence a ecting reasoning and decision making. To cope with this, several solutions are possible, e.g. areas can be monitored either with recurrent human inspections or by deploying static sensors regularly streaming information. Besides being expensive in time, resource and maintenance, these solutions are neither exible (e.g. sensors are bound to speci c locations) nor scalable, especially in large scale scenarios such as urban environments.
Works in Robotics have demonstrated that using an autonomous mobile agent as a moving sensor is a valid alternative solution, but have rather focused on achieving robotics tasks (perception, planning, navigation, etc.) with high precision [?]. We focus instead on integrating robots as part of a larger knowledge-based system, whose role is to store and manage the information that robots collect, share and revise continuously for the achievement of their daily activities. As semantic technologies have proven to be successful in contexts where knowledge from heterogeneous sources needed to be integrated to enable a variety of light-weight applications [?], our question here becomes can we use the available semantic technologies to deploy applications where robots act both as data collectors and data consumers for a centralised knowledge base?
To answer this question, we implemented the HanS system, where a mobile robot is deployed as part of a central knowledge-based system to autonomously assess the correct compliance with the Health&Safety rules holding in the Knowledge Media Institute (KMi). We focus on rules concerning the use of appropriate signage for emergency appliances and runaways, e.g. \Are re extinguishers clearly labelled? " (Rule01), or the presence of forbidden objects in restricted areas, e.g. \Are electric heaters away from con ned areas? " (Rule02). This demo presents the implementation of the system and a simulation of HanS at work, showing users how robots can be integrated in simple applications through combining the available semantic and robotics technologies. 2
HanS is implemented as a modular architecture with the idea that each module can be easily replaced with more advanced implementations depending on platforms and tasks. As shown in Figure 1, the system is articulated in (i) the Knowledge Component, that manages the knowledge level, and (ii) the Sensing Component, performing monitoring and data collection.
The Knowledge Component is composed by three modules, i.e. the Knowledge Base, the Triplestore, and the RESTful Server. The Knowledge Base (KB) module contains all data necessary for the system to reason over H&S rule violations { namely, information about the physical environment, called Semantic Map [?], and the de nition of the H&S Rules. The semantic map describes objects and areas in terms of their position w.r.t. the geometrical map used by the robot for its localisation and navigation, as well as other objects/areas' properties such as types and names. For the sake of simpli cation, rules are encoded as constraints or restrictions about speci c objects in speci c locations, following a baRESTful Server
Triplestore Knowledge Base H&S Rules
Semantic
Map
KB Interface Standard ROS
Modules Inspection Routine Object Detection sic schema h:RestrictionRule,:hasForbiddenObject,:Objecti. h:RestrictionRule,:hasLocation,:Areai. For example, a :HeaterRestriction establishes that a :Heater cannot be located in a space of type :Activity1. The knowledge base is managed by the Blazegraph triplestore2, which natively supports both geospatial reasoning and inference checking. The RESTful server, implemented in Blazegraph as a simple Servlet, is the module in charge of communication between a user, the knowledge base and the robotics platform.
The Sensing Component is in charge of collecting information and includes the physical robot and all the modules to operate it. The platform we use is a Turtlebot 2 equipped with a Hokuyo Laser Range nder (for localisation and navigation), and an Orbbec Astra RGBD sensor for object detection. All the software modules, managing speci c functionalities of the robot, are implemented through the Robot Operating System3 (ROS) framework using a combination of custom and standard ROS modules. For example, we use move base for navigation and kobuki node for motor control. A custom Object Detection module is used for the task of detecting an object and its position when navigating an area. This is currently implemented as an ARTag4 detection process, associating an ARTag with an object class (e.g. :Heater), but can be easily extended with more sophisticated systems. An Inspection Routine module then implements the logic of the exploration process necessary to check a speci c rule: for example, the module will instruct the robot to perform a 360 -spin when reaching an area (Rule02), or to perform an inspection along a wall with a speci c height (Rule01). Finally, the Behaviour Manager uses behavioural trees [?] to activate a speci c robot behaviour depending on the type of rule to be checked (currently activated upon user requests, e.g \check for Rule02"), while the KB interface manages the communication with the server. 1 Activities are portions of open space in KMi. 2 https://www.blazegraph.com/ 3 http://www.ros.org/ 4 https://en.wikipedia.org/wiki/ARTag
The demonstration will show HanS in action using recorded videos and, if conditions allow, a simulation of the robot and environment. Users will be shown the whole system in-depth, and will be able to check live for a wider range of rules. Discussion on possible extensions (e.g. reasoning through SPIN5 rules, object detection through neural network approaches etc.) will also be held. A demo of the process is currently available online6, described as follows.