**1.4. Data aggregation in multi-robot sensor networks**

72 Wireless Sensor Networks – Technology and Protocols

communication path to the reference robot can be established.

monitoring (intelligent transport system) etc have also much potential.

**1.3. Open problems in multi-robot sensor networks** 

**1.2. Applications of multi-robot sensor networks** 

sometimes robots are out of the network area so that break the network connectivity. Hence, how to keep the connectivity all the time is a crucial issue so that it already became a hot research topic (Mi et, al., 2010). Besides, the reliability and robustness as well as secure et al. are some other concerned research topics. With respect to communication techniques in MRSN, a network with the goal of search and rescue is described in (Reich & Sklar, 2006). In this paper, they proposed an entirely distributed gradient propagation (GP) algorithm. Each sensor in the paper independently executes the GP algorithm and broadcasts after some independent, randomly chosen time interval. The robots sensors estimate their target by "hot" values and "cold" values, where the "hot" values become the searching target. In (Sheng, et al. 2006), for reliability and robustness, a distributed biding algorithm was proposed for multiple robots in exploration tasks to address the problems caused by the limited communication range. In this algorithm, all the robots work asynchronously. There are three states for each robot that (1) sensing and mapping, (2) bidding and (3) traveling. A distributed algorithm that makes mobile robots in a multi-robot system aware of network connectivity was discussed in (Leyzx, et al., 2009). The basic idea is to take a "fixed" robot as the reference robot that keep in touch with at least one neighboring robot from which a

Due to its flexibility, operability, mobility and self-organization, the applications of MRSN has been increasing (Maxim & Gaurav, 2005), (Trigui S, et, al., 2012). Harsh environmental monitoring is the most popular application of MRSN, for example let wireless robots get into Amazon rainforests where it is very dangerous for human get inside or let them climb to Mount Everest where there is not enough Oxygen for human and covered by snow all over the mountain. In medical application, if the MRSN can help the nurses to do some simple task like checking body temperature and sending to a doctor, which would save much more labors in some countries those short of nurses. One of the most important utilizations of MRSN is that it can be used to detect nuclear radiation and to accomplish some other relevant tasks. The most recent example is the Fukushima nuclear leakage where if a MRSN was applied, it would have alleviated damage. Some other applications like outer space monitoring (space junk detection), industrial monitoring (quality control), disaster monitoring (forest fire detection), agriculture monitoring (soil moisture detection), traffic

In MRSN, when an event occurs, multiple robots in the near area sense the event data and generate an abundance of sensed data; however, many of the data generated in the same area are highly redundant. Hence the transmission and relaying of all generated data caused a big waste of bandwidth and energy; it also causes data collision and congestion so that result in low efficiency of data gathering. On the other hand, similar to wireless sensor network, WRSN could not avoid the shortcoming of lack of continuous energy supplement. We focus on data aggregation technology for collecting data in MRSN. Data aggregation (Rajagopalan & Varshney, 2006) is a process of aggregating the data from multiple robot sensors to eliminate redundant data and provide fused information to the base station. Considering from the point of data redundant, data aggregation can collect the most efficiency data. However, transmission delay and data accuracy are also important in many applications such as military application and architectural application. Hence trading off transmission delay, energy consumption and data accuracy is an important issue. There are several typical algorithms of data aggregations. PEGASIS (Lindsey & Raghavendra, 2001) is one of energy efficiency chain based data aggregation protocols that employs a greedy algorithm. The main idea of PEGASIS is forming a chain among the sensor nodes so that each node receives (or transmits) fused data from (to) the closest neighbors. The data gathered are sent from node to node, and all the sensor nodes take turns to be the leader for transmission to the Base Station. Data Funnelling (Petrović, et. al, 2003) is another scheme that sends a stream of data from a group of sensor readings to destination. Moreover, they proposed a compression method called "coding by ordering" to suppress some readings and encoding the values in the ordering of the remaining packets. On the other hand, LEACH (Heinzelman W., 2000) is one kind of energy saving schemes in which a small number of clusters are formed in a self-organized manner. A designated sensor node in each cluster collects and combines data from nodes in its cluster, then transmits the result to the BS. Directed Diffusion (Intanagonwiwat, et. al, 2000) is a kind of data centric routing protocols. The sink broadcasts an interest message to all the sensor nodes, and the nodes gather and transmit the sink-interested data to the sink. When the receiving data rate becomes low, the sink starts to attract other higher quality data.

Regarding the trade-offs, (Boulis, et. al, 2003) proposed an energy-accuracy tradeoffs algorithm for periodic data-aggregation which is a threshold-based scheme where the sensors compare their fused estimations to a threshold to make a decision of regarding transmission. Energylatency tradeoffs algorithm (Yu at. el., 2004) is proposed for minimizing the overall energy consumption of the networks within a specific latency constraint where data aggregation is performed only after a node successfully collects data from all its children and its own local generated data. ADA (Adaptive Data Aggregation) (Chen et. al., 2008) is an adaptive data aggregation (ADA) for clustered wireless sensor networks. In ADA, sensed data are aggregated on two levels; one is aggregated at sensor nodes controlled by the reporting frequency (temporal reliability) of nodes; another is aggregated at cluster heads controlled by the aggregation ratio (spatial reliability). The reliability of observed data that is decided by the number of arrival data at sink node is compared with the reliability of desired data, which is decided by the application. According to comparison, nine characteristic regions and nine states are defined in which the eight states must change into the desired state through the calculating and adjusting of observed reliability.

Most of the previously mentioned works focus on energy saving and aggregate as much data as possible. As a result, they prolong the transmission delay. Many works aimed to achieve energy-delay trade off, however they still have shortcomings for example (Yu at, el., 2004) has long waiting time at nodes with less event data while the constant latency makes the networks very inflexible in (Galluccio L. & Palazzo S., 2009). A desired energy-delay tradeoff is achieved in (Ye Z. et al., 2008); however the algorithm ignored the issue of data accuracy. Energy-delay-accuracy tradeoffs in (Mirian F. & Sabaei M.) and (Chen et al., 2008) adapt to a situation that could be described by the following question: 'what is the average temperature of this area at this hour?' The algorithms did not consider delay and accuracy among nodes and data, which may lead to large data deviation as well as transmission delay in some other applications.
