Cyber-physical systems bridge the cyber world (e.g., information, communication, and intelligence) to the physical world through lots of sensors and actuators. A CPS may consist of multiple static/mobile sensor and actuator networks integrated under an intelligent decision system. For each individual WSN, the issues such as network formation, network/power/mobility management, security, etc. would remain the same. However, CPS is featured by cross-domain sensor cooperation, heterogeneous information flow, and intelligent decision/actuation. In this section, we will review some CPS applications and make observations from these prospects. The concept is illustrated inFig. 3. For example, a CPS can facilitate greenhouse asset management through the deployment of multiple WSNs [97]. Each WSN is composed of multiple sensors and actuators to form a climate control system with lighting, cooling, heating, carbon dioxide generating, watering, and fertilizing subsystems. Thus, light intensity, temperature, humidity, and density of carbon dioxide need to be collected and reported. The decision system will transform these sensing data into high-level knowledge (e.g., the proportion of each type of fertilizers) to trigger actuators to maintain good environmental factors in the greenhouse. Note that multiple actuations may coexist (e.g., the cooling subsystem may cowork with the carbon dioxide generating subsystem). Efforts on sensor–actuator and actuator–actuator coordinations can be found in [98,99]. Robotic actuators are introduced in [100].
Below, we review some CPS applications and platforms.Table 3gives a summary of these efforts.
WSN Mobile sensing networks User
Cellular/WiFi stations
actuator sensor
Sensing Data Actuation
Knowledge
Storage/applications
platforms computing/intelligent
decision systems
User User
Fig. 3. A CPS architecture model.
Table 3
Comparison of CPS applications & platforms.
Reference Functionality & application Sensor type Heterogeneous information flows
[101] Physical rehabilitation & telemedicine ECG, EMG, EEG, SpO2, accelerometer,
& tilt sensors
BSNs/GPRS/Bluetooth/WLAN
M-Health [102] Security & telemedicine ECG & PPG BSNs/Bluetooth/GPRS
ANGELAH [103] Elder assistance Video camera, audio, RFID, & smart door lock
WSN/WiFi
[104–107] Emergency navigation Light, smoke, & temperature sensors WSN
FireGrid [6] Emergency navigation Temperatures,heatflux,
&gas(O2,CO,&CO2)
WSN/Internet
CarWeb [108] Traffic monitoring & mining GPS 3G/WiMAX/Internet
Nericell [109] Traffic monitoring Accelerometer, microphone & GPS, GSM/GPRS/EDGE/Internet
VTrack [110] Route planning GPS & WiFi 3G/WiFi/Internet
[111] Transit tracking Accelerometer & GPS 3G/Internet
BikeNet [112] Experience sharing among bikers Accelerometer, microphone, GPS, magnetometer, CO2, & Temperature
WSN/GSM/GPRS/Bluetooth/WiFi /Inetrnet
ParkNet [113] Parking Ultrasonic & GPS Cellular/WiFi/Internet
[114] Traffic light control RFID WiFi
CenceMe [12] Social networking Accelerometer & GPS Cellular/WiFi
[115] Social networking Camera & accelerometer WiFi
Micro-Blog [116] Social networking GPS, WiFi, & GSM WiFi/GSM/cellular
Video game [117] Video game Accelerometer & compass BSN
[118] Social game Accelerometer & compass BSN/WiFi/3G
4.1. Health-care applications
Cross-Domain Sensing: Future health-care applications would benefit from wearing small sensors by patients [119].
ECG (electrocardiogram), EMG (electromyography), EEG (electroencephalography), SpO2, accelerometer, and tilt sensors are adopted for physical rehabilitation in [101]. Ref. [102] considers security issues in telemedicine when multiple body sensor networks (BSNs) coexist. The system uses ECG and PPG (photoplethysmogram) sensor and designs an authentication mechanism to verify whether two nodes belong to the same BSN through the timing information of users’ heartbeats, which may uniquely represent an individual. Ref. [103] proposes an innovative elder assistance framework, termed AssistiNG ELders At Home (ANGELAH), which probes passing-by volunteers to provide prompt help to elders in emergency situations.
Heterogeneous information flow: The rehabilitation system in [101] has a 3-tier network architecture to monitor the physiological signal of remote patients. The lowest tier is a ZigBee body sensor network connecting to ECG, EMG, EEG, etc.
The middle tier contains GPRS/Bluetooth/WLAN connecting to personal devices (e.g., PDA, cell phone, and PC), which serves as sinks for local BSNs. The upper tier is a broadband network connecting servers to record and analyze collected data from individual patients and issue recommendations, if necessary. In [101], heterogeneous networks are used to connect video cameras, sound sensors, RFID readers, and smart appliances (e.g., smart door lock). Clearly, multi-tier networks are widely used in CPS applications.
Decision/actuation system: A recommendation system is needed in [101] for doctors and nurses. In ANGELAH [103], there are six roles: Sensing Entity (SE), Actuator Entity (AE), Home Manager (HM), Surveillance Center (SC), Local Responder (LR), and Locality Manager (LM). SEs are sensors deployed in elders’ living spaces. AEs are actuators set up in elders’ living spaces and controlled by HM. For example, cameras in a living room are AEs, which can be triggered by RFID readers when elders’
tags are detected. HMs are local servers at homes for gathering sensing data and controlling AEs. SCs are responsible for coordinating prompt responses and selecting the most adequate volunteers to provide help. When an emergency situation is detected, HM informs SC the emergency level, type, cause, and the kind of assistance that may be needed. LRs are nearby volunteers who are willing to provide prompt help. ANGELAH defines the proximity of LRs by the coverage of network access points. It is worth noticing that multiple elders may need assistance in the same network coverage.
4.2. Navigation and rescue applications
Cross-domain sensing: We define a responsive navigation/rescue system as one which takes real-time sensing observations into consideration. Ref. [104] considers navigating people in a dangerous region with multiple emergency points and one safe exit. Guiding people in fire emergency to safe exits in 2D/3D environments is discussed in [105,106], where a 3D environment involves multiple floors with stairs. Such systems all need smoke, temperature, and/or humidity sensors.
For rescuers, infrared, smoke, camera sensors, and life detectors may be needed [120,121]. Note that rescuers also need navigation services to find lives. The Robobees project [122] intends to develop micro air-robotic insects to facilitate search and rescue in unreachable areas.
Heterogeneous information flows: To find safe navigation paths without passing any obstacle, [104] relies on exchanging attractive and repulsive potentials among sensors and exits. Similarly, exchanging ‘‘altitudes’’ is used in [105,106].
Navigation without location information is discussed in [107], where sensor nodes exchange information cooperatively to identify the medial axes of danger areas. Load-balanced navigation is addressed in [123]. These applications all need lots of sensor-to-sensor, sensor-to-human, and sensor-to-infrastructure communications. While WiFi/ZigBee/IEEE 802.15.4 is commonly assumed in such systems, more complicated networking scenarios are demanded.
Intelligent response system: In next-generation large-scale navigation systems with emergency response and evacuation supports, both scalability and fault tolerance are important factors. Agent-based approaches allowing computation tasks to be executed in distributed and parallel manners are studied in [124]. For emergency support, FireGrid [6] adopts high performance computing grids. FireGrid has four major components: (1) data acquisition and storage component, (2) simulation component, (3) agent-based command-control component, and (4) Grid middleware component. In particular, there is a Data Grading Unit to filter and validate accuracy and reliability of sensing data. The simulation component has a set of computational models to interpret the current status and predict the future status of fires. The computing grids provide access to processing units in parallel by emergency tasks. To support decisions for emergency responders, there are a query manager agent and a set of Command, Control, Communication and Intelligence (C3I) user-interface agents to interact with users. Ref. [125] designs a mechanism to avoid navigation congestion and provides firemen rescue commands to intelligently eliminate key dangerous areas such that the number of trapped people can be reduced.
4.3. Intelligent Transportation Systems (ITSs)
Location tracking and road information sensing: Transportation is key part of our daily life. ITSs are important applications of CPS [126]. Many ITSs projects [127,128] have been developed. Needless to say, GPS information is a must in ITSs. In [109], accelerometers and microphones in mobile phones are used to annotate traffic conditions in a map, such as braking, bumping and honking, thereby allowing us to search for driving directions with less stress, such as chaotic roads and intersections.
Considering GPS outage, [110] uses WiFi interfaces on mobile phones to track nearby APs and their centroids for positioning.
To track the locations of public vehicles (e.g., buses), [111] proposes a cooperative model between GPS and accelerometers.
BikeNet [112] exploits inertial sensors and air sensors for bikers’ experience sharing. Several projects consider parking applications through GPS, ultrasonic, and ranging sensors [129,130,113].
Heterogeneous networking: OBD [131] and CAN-Bus [132] have been developed for intra-vehicle communications.
DSRC/WAVE [133] are defined for V2V and V2R communications. ETC systems [134] are widely deployed for toll collection.
Through inter-vehicle communications, rear-end collision prevention is discussed in [135,136]. By collecting real-time car locations, CarWeb [108] shows how to mine the current traffic conditions of road segments including waiting time at intersections.
Monitoring and provisioning systems: It is still a difficult task to accurately measure traffic conditions. VTrack [110] uses mobile phones to help estimate traffic delays, detect hot spots during rush hours, and provide real-time route planning with sparely-sampled GPS data. It applies an intelligent hidden Markov model to estimate vehicle trajectories and eliminate
Fig. 4. Problem spectrum in cyber-and-physical world using Tai-Chi exercises as an example.
outliers through map matching and to associate with each sample the most likely road segment. ParkNet [113] exploits sensing capability of drive-by vehicles to detect the occupancy of parking spots. If lane-level vehicle information can be accurately measured, [114] develops an intelligent traffic signal control protocol to speed up car-crossing at intersections.
4.4. Social networking and gaming applications
Sensor-enhanced interaction: Traditional social networks are Web-based (e.g., Facebook, Windows Live, and YouTube).
CPS-based social networking would involve more physical inputs to enrich interactions among users. CenceMe [12] allows users to share their sensing information via mobile phones (e.g., audio, motion, acceleration, and location). Therefore, users’
moods, activities, mobility patterns, locations, etc. become exchangeable information in the cyber world. Ref. [115] shows an interesting way to record and share water-drinking behaviors as socialization tools. Through sharing locations, Micro-Blog [116] encourages social participants to share and query information with each other via smart phones. In the area of interactive games, adopting physical inputs, such as those from inertial sensors, is already popular. Information processing of inertial sensors in BSNs has been intensively studied [137–139]. In [117], a CPS-based video game is presented. A social network where participants can wear inertial sensors and practice traditional Chinese Tai-Chi over the Internet is presented in [118]. This platform allows users to share conventional information as well as sensory data in a real-time manner. InFig. 4, we show the spectrum from ‘‘virtual’’ Tai-Chi, to ‘‘cyber–physical’’ Tai-Chi, and then to ‘‘physical’’ Tai-Chi.
Networking scenarios: To collect sensing data, BSNs/WSNs are needed. Participants then use the Internet or 3G networks to interact with each other. The use of networks should be context-aware. For example, Micro-Blog [116] tries to balance between localization accuracy and battery lifetime. A mobile phone switches between two localization modes, WiFi and GPS modes. WiFi mode is the default mode if the WiFi fingerprint does not change over time. Once it detects movement over a threshold, it switches to the GPS mode.
Socialization and interaction platform: A social platform needs to connect many users around the world, thus demanding huge data processing and storage capability. CenceMe [12] uses a split-level classification to process raw sensing data. A user’s mobile phone first classifies data into high-level information, called primitives. CenceMe provides an audio classifier and an activity classifier to recognize human voices and activities (including sitting, standing, walking, and running), respectively.
When primitives arrive at the backend server, five complex classifiers, namely conversation classifier, social context, mobility mode detector, location classifier, and am I hot, are designed to retrieve second-level information, called facts, to reflect users’
social behaviors, such as on-going conversations, nearby buddies, user’s mobility pattern (e.g., if he/she is in a vehicle), and location type (e.g., restaurant, library, etc., which is obtained via Wikimapia GIS databases [140]). Interestingly, the ‘am I hot’
classifier can recognize users’ social stereotypes (e.g., party animal, cultured, and healthy). Facts are stored in a database for retrieval and publishing. In Micro-Blog [116], a user can query a specific region. If there exists information in a database matching the query, the server will reply to this query. Otherwise, the server will select those phones located in this region and forward this query to them for possible responses. To encourage responses to socialized queries, a give and take scheme is adopted, where users earn/pay credits when they reply to/issue queries.