iConfig: What I See is What I ConfigureMichael Haus
Technical University of MunichAaron Yi Ding
Technical University of Munich
Pan HuiHong Kong University of Science and Technology
Jorg O�Technical University of Munich
ABSTRACTManaging IoT devices in urban areas is becoming crucial be-cause the majority of people living in cities and the numberof deployed IoT devices are steadily increasing. In this demo,we present iCon�g, an edge-driven platform dedicated tomanage densely deployed IoT devices in smart cities. Ourgoal is to address three challenging issues in current IoTmanagement: registration, con�guration, and maintenance.�e core of iCon�g is its programmable edge module, whichcan be deployed across smartphones, wearables, and smartboards to con�gure and interact with physically proximateIoT devices. Our system evaluation shows that iCon�g cane�ectively address the aforementioned IoTmanagement chal-lenges by harnessing mobile and edge cooperation. �e livedemonstration further showcases how iCon�g can utilizespeech recognition on smart wearables (e.g., smart glass) toachieve more natural and frustration-free IoT management.
1 INTRODUCTION�e majority of people lives in cities and the number ofdeployed Internet of �ings (IoT) devices are steadily incre-asing. Managing urban areas and its applications is hencebecoming important. �ese urban IoTs support the smartcity concept [1] which integrates traditional and moderninformation and communications technology (ICT) for a uni-�ed and simple access to services for the city administration(e.g., managing urban areas) and the residents [2, 3]. �e aimis an enhanced use of public resources, improving qualityof services for citizens while reducing operational costs ofpublic administration [2]. In spite of a growing demand forIoT management to keep an up-to-date overview of denselydistributed devices during di�erent phases of their life cycle,including installation, registration, user customization, anddevice control. �ere is a lack of instrument to seamlesslymanage large IoT deployments in which ad hoc managementis becoming untenable, especially for IoT devices withoutInternet connectivity. �erefore, we propose iCon�g, anedge-driven platform to bridge such gap. Fig. 1 shows iCon-�g in the context of smart environments taking advantage ofwearable and edge computing in the form of programmabledevices to run edge modules on smart glasses or IoT boards.�is allows iCon�g to consolidate various connected devices
WebserviceControl
Interface
Config Storage
Backend
Edge Module
IdentifyRegisterUpdateUpdate
...Action Queue
Sync
hron
izat
ion
Co
mm
un
icat
ion
Inte
rfac
e
Configuration
Smart GlassSmart WatchSmartphone
Smart Environment
BLE Beacon
iConfig
IoT Boards
Add-on Servicesfor BLE Beacons- Monitoring- Debugging- Localization
Figure 1: iCon�g in the context of smart environ-ments
to its management backend, which enforces a uni�ed con�-guration procedure. As an example, we successfully testediCon�g on Android smartphone and smart glass [4]. In par-ticular, iCon�g edge module is dedicated for users interestedin managing and interacting with IoT environments. �edesign principles of iCon�g include: 1) automatic con�gura-tion of IoT devices to avoid miscon�gurations which becomeone of the dominant causes of system failures [5], 2) easy touse frontend, 3) orchestrate devices via global view, and 4)serve as platform for developer to enable add on services.In this demo, we use iCon�g to con�gure Bluetooth Low
Energy (BLE) beacons, which represent one of the most chal-lenging classes of IoT devices due to missing internet con-nectivity. �ese ba�ery powered BLE beacons are small-sizewireless devices that transmit a short-range BLE signal tomobile computing devices (e.g., smartphones) [6]. Via BLE,users’ devices are noti�ed of the beacon proximity by recei-ving signals which contain contextual information, typicallyabout indoor surroundings and its contents. �us, the re-ceiving end is able to perform location aware actions, suchaccessing speci�c URLs for marketing and assistance pur-poses [7]. �e ability of iCon�g to programmatically adjustdevice parameters facilitates di�erent use cases: 1) set upan automated testbed for research projects, 2) debug andmonitor IoT devices, and 3) energy aware management ofIoT devices deployed in smart buildings.
Figure 2: Smart glass with BLE beacons, currently con-�gured beacon blinks
2 ICONFIG FOR IOT MANAGEMENT�e iCon�g system architecture, as illustrated in Fig. 1, con-sists of twomajor modules: mobile edge module and backendmodule. �e framework is able to identify, register, and up-date IoT devices (as in our case BLE beacons). For instance,supported by our speech recognition functionality, a userwearing an iCon�g-enabled smart glass can discover, regis-ter, and con�gure BLE beacons while walking around. Fig. 2shows the demo hardware including smart glass and BLEbeacons. In the discovery phase, the edge module showsonly unregistered beacons. A�er discovering unregisteredbeacons, the user is able to register one beacon at a time.�e beacon shows a red light as feedback for device identi�-cation. When the beacon identi�cation was successful, theuser can register the beacon to the iCon�g backend with ad-ditional information for device localization, such as nearestroom number, picture of device place. A�erwards, the edgemodule will automatically con�gure the BLE beacon with adefault con�guration including password, URL, transmissionpower, and packet advertisement rate. Finally, edge modulewill synchronize all con�guration data to the backend. �eregistration has to be done only once per beacon. For ad-vanced management, iCon�g backend provides a controlinterface for the administrator, including a global view aboutinstalled BLE beacons. �e device update is independent ofuser actions and automatically triggered when two condi-tions are satis�ed: 1) adapted BLE con�guration from theadministrator via iCon�g backend available, and 2) currentlydiscovered beacon by the user via iCon�g edge module.
3 EVALUATIONWe analyzed the system performance of iCon�g regardingmemory usage and scalability over multiple beacons. Additi-onally, we conducted a user study to highlight the drawbackof manual con�guration, which is untenable in dense deploy-ment.Regarding the memory usage of iCon�g edge module,
we measured a deployment on Android smartphone (OS7.1.1) in o�ine and online mode during con�guration of tenBLE beacons. In online mode, the edge module used 6.50
Table 1: Beacon con�guration time (s) for scalabilitytest from one to ten beacons
1 2 3 4 5 6 7 8 9 10
2.7 5.3 7.5 10 12.4 15.1 16.5 16.4 17.8 22
± 0.98MB similar to o�ine mode with a memory usage of6.47 ± 0.96MB. �ese results show that the memory foot-print of the iCon�g edge module is small enough to run onprogrammable IoT devices.For scalability we evaluated ten cases (from one to ten
beacons) each over 20 rounds in a dense testbed deploy-ment. Our evaluation yielded 18 unauthenticated connecterrors (meaning BLE con�guration is not possible) out of1100 connect a�empts, i.e., rate of 1.64 %. Table 1 shows alinear increase of con�guration time over all beacons, onaverage an increase of 2.2 s per beacon. In addition, we evalu-ated the success rate which describes whether con�gurationparameters were correctly set. �e lowest rate was 75 % du-ring con�guration of four beacons. In most cases, iCon�gachieved 100 % success rate.Our user study further revealed the gap between a ma-
nual and automatic con�guration system. Ten participantsmanually con�gured three beacons with a prede�ned con-�guration using the vendor application. In the next step,the same beacons were con�gured via iCon�g, in which thedefault con�guration was automatically wri�en to each BLEbeacon. �e manual con�guration took in average six timeslonger than iCon�g automatic con�guration, which re�ectsa time saving of 83 %. Only 1/3 of all manual con�gurationswere entirely correct. On the other hand, iCon�g achievedfor all BLE con�gurations the success rate of 100 %.
4 HANDS-FREE IOT DEVICECONFIGURATION
We take advantage of speech recognition to enable handsfree device con�guration and implemented two prototypesfor di�erent device types: smartphone and smart glass. Onthe smartphone the speech recognition can be optionallyactivated, on the smart glass it is automatically activated toallow a convenient usage of the iCon�g edge module. �espeech recognition is based on two implementations: Googlespeech recognition and Pocketsphinx [8], and works comple-tely o�ine on the smart glass. iCon�g can be activated viathe key phrase geronimo for device discovery and registra-tion. A�erwards, the user is able to control iCon�g via thefollowing speech commands: select nearest beacon automa-tically detected via strongest received signal strength, targetspeci�c beacon via beacon ID, identify as visual feedback ofthe beacon, and register beacon at iCon�g backend.
2
To capture an image of beacon placement for easier loca-lization, we implemented a customized camera dialog viaspeech control. �e keyword shot takes an image and thezoom range of the camera is divided into �ve levels. Userscan control the zoom range via the speech commands: more,less, min, and max.
To enhance the robustness of our speech recognition (e.g.,number three sometimes recognized as “free”), we used se-veral metrics for string similarity of the speech input: Eucli-dean, Levenshtein, and Jaro-Winkler distance. In our se�ingthe Jaro-Winkler distance, which is designed for short stringsand can detect typos, provided the best results to calculatethe string similarity of iCon�g speech input.
5 DISCUSSION AND OUTLOOKA key observation from our user study and experiments isthat the interaction among users, their smart gadgets and sur-rounding IoT devices via the conventional screen-keyboardsetup is far from optimal. Especially for smart cities with amultitude of services empowered by IoT devices, the systeminteraction should be more natural and �uent. �is is themain reason for our prototype dedicated for smart wearables(e.g., smart glass). �e combination of speech recognitionand hands-free devices can enable a more integrated inte-raction during user movement and limit the distraction ofuser a�ention. iCon�g is hence our endeavor to enhanceuser experience and to streamline the management of largescale IoT deployments, especially for low-budget devicessuch as BLE beacons without backend connectivity.
To tackle more challenging scenarios of IoT managementwhere IoT devices are deployed in harsh environments (e.g.,high ceilings, steel-making factories), we plan to exploreusing iCon�g-enabled drones for achieving autonomous con-�guration, which is highly desired in dense IoT deployments.
6 DEMONSTRATION PROCEDUREIn this demonstration, we present a typical use case of ourplatform to con�gure BLE beacons via smart glass overspeech control. For comparison, we provide a smartphoneto con�gure the beacons. �e testbed layout is depicted inFig. 3. �e laptop runs the iCon�g backend server togetherwith a MongoDB database which is accessible via a local Wi-Fi hotspot. �e iCon�g backend server receives the beaconcon�gurations and enables add-on services. Besides that, thelaptop projects the screen content of the smart glass to alarger monitor. �is monitor shows the GUI of the iCon�gedge module running on the smart glass and the controlpanel of the iCon�g backend. For device con�guration, ourdemo includes a smart glass and a smartphone running theiCon�g edge module to identify, con�gure, and update BLEbeacons. We provide ten BLE beacons for con�guration.
Figure 3: Demo setting
Our demo includes the following devices: 1) smart glassrunning iCon�g edge module with speech control for hands-free device con�guration, 2) smartphone running iCon�gedge module for conventional manual input or optionalspeech control, 3) ten BLE beacons for device con�gura-tion, 4) a laptop to run the iCon�g backend server and toproject the screen content of the smart glass to an externalmonitor, and 5) a headphone for improved speech control innoisy environments.We would need space for a poster board and a table, big
enough for one laptop and a 24-inch monitor. Our prototypesystem requires approximately 30 minutes to setup.
In addition, we need a table and power for the demo. Wealso prefer to have a large monitor with HDMI or VGA cableto show the screen content of the smart glass running theiCon�g edge module. We will cast the control panel of theiCon�g backend module on the monitor as well.
REFERENCES[1] Scha�ers et al. Smart Cities and the Future Internet: Towards Coope-
ration Frameworks for Open Innovation. In Future Internet. LectureNotes in Computer Science, Vol. 6656. Springer. 2011. 431–446.
[2] Zanella et al. Internet of�ings for Smart Cities. IEEE Internet of �ingsJournal 1, 1 (2014), 22–32.
[3] Zdraveski et al. ISO-Standardized Smart City Platform Architectureand Dashboard. IEEE Pervasive Computing 16, 2 (2017), 35—43.
[4] h�p://madgaze.com/x5/ (visited on 05/20/2017)[5] Xu et al. Systems Approaches to Tackling Con�guration Errors. ACM
Computing Surveys 47, 4 (2015), 1–41.[6] Wang et al. Managing Large Scale, Ultra-Dense Beacon Deployments
in Smart Campuses. In Proceedings of IEEE INFOCOM WKSHPS 2015.[7] h�p://www.airport-world.com/news/general-news/5976-lax-uses-
bluetooth-beacon-technology-to-improve-wheelchair-operations.html (visited on 05/20/2017)
[8] D. Huggins-Daines et al. Pocketsphinx: A Free, Real-Time ContinuousSpeech Recognition System for Hand-Held Devices. In Proceedings ofthe IEEE ICASSP. 2006. 185–188.
3
