Whenever an alteration is sensed in the environment, a task allocation process needs to be performed between the robots and the tasks. In this case, the task allocation process is considered to be a dynamic process. The centralized approach can be considered as a fast solution to this type of problem. However, if a quite preferable solution is needed, a distributed assignment technique will be better. No centralized leader or controller is needed in this approach, and that is why tasks‟ scheduling in swarm robotics must be performed because of this distribution process. However, one of the main disadvantages, that result from this decentralization is that problem complexity is increased, because the robot view in the environment becomes limited.
4.2.1 Centralized approach
In centralized approach, there is a leader or central unit which is responsible for tasks assignment to the robots. Gigliotta et al. [27] have evolved a dynamic task allocation rules by communicative interactions in a group of homogeneous robots. They focused on the development of a team of robots in which one and only one individual robot (the ‟leader‟) must differentiate its communicative attitude from that of all the others (‟non-leaders‟). The robots evolve their capabilities to distinguish their roles by the discrimination of their signals. The leader robot has to maximize the value of its communicative signal, while all other robots have to minimize their signals‟ value. The leader robot tends to send high signal‟s value, while the non-leaders tend to send a low signal‟s value. The fitness of a group of robot was calculated in the following way. The average of the differences between the output signal of the current leader which has the maximum value and the output signals of all other non-leader robots was calculated every iteration. A number of trials need to be implemented. At the end, the average of the calculated value for all iterations of all the trials can be considered as the fitness value.
35
Formally, this is how fitness value was calculated:
𝐹 = 𝑀𝑎𝑥 − 𝑂𝑖 𝑁 𝑖 𝐶 𝑗 𝐶(𝑁 − 1) , Equation (4-17)
where the number of robots in each group represented by N, such as 10. While, the total number of iterations of each individual called C, such as 1000 iterations * 40 trials = 40000. Also, the signal‟s value of the current leader is Max and the signal‟s value of robot j is Oj.
4.2.2 Distributed approach
In distributed approach, there is no central unit to take care of the task allocation. So, each robot in the swarm has to identify the task it must perform. Brutschy et al. [13] have developed a self- organized method for allocating the individuals of a swarm to tasks that are sequentially interdependent. The proposed method does neither rely on global knowledge nor centralized components. Moreover, it does not require the robots to communicate.
De Mendonça et al. [16] have proposed a simple yet efficient distributed control algorithm to implement dynamic task allocation in a robotic swarm, in which every robot in the swarm has to identify the task it must perform. The swarm task assignment is represented by A = {a1; a2;…; aρ}, wherein aj identifies the task assigned to robot identified by idj. Note that given a task assignment, say A, it is possible to set counters associated with it, which denoted as CA, as described in the following equation:
𝐶𝑗 = 𝐶𝐴 𝑡𝑗 = ∅ 𝑎𝑟, 𝑡𝑗 , 𝑝
𝑟=1
36
Wherein function ∅ is defined as follows:
∅ 𝑎𝑟, 𝑡𝑗 = 1 𝑖𝑓 𝑎 = 𝑡;0 𝑜𝑡𝑒𝑟𝑤𝑖𝑠𝑒
Equation (4-19)
Thus, at the end of the construction, the counter set CA would reflect the distribution of the robots as assigned to each task and defined by assignment A. Solving the dynamic task allocation problem consists of finding the task assignment A* = {a*1; a*2;…; a*ρ}, that verifies:
∀ 𝑡𝑗 ∈ 𝑇 𝑎𝑛𝑑 ∀ 𝑝𝑗 ∈ 𝑃.
Equation (4-20)
Keshmiri and Payandeh [12] have presented the percentile values of the distributional information of the tasks to reduce the task space into a number of subgroups that are equal to the number of robotic agents.
Liu et al. [17] have enhanced an optimal assignment method with a decentralized leadership. They have proposed an algorithm in which local search processes are executed simultaneously without any connections among these processes. In addition, this technique is totally decentralized in which messages‟ broadcast and a multi-hop communication would never be used.
Zhang et al. [28] have established hierarchical allocation architecture for the set of robots in the population. Two algorithms are implemented in this hierarchy. First, a simple self-reinforcement learning model is utilized in the top level of the hierarchy which turned the initially identical particles into specialists for distinctive task types by using the social insects‟ method. This method gives a robust and flexible labor‟s division as can be seen from figure 4.1. However, when the single type of task is considered in the lower level of the hierarchy, the Ant-colony technique is implemented to solve the task assignment issue. A local communication technique is used to share information among robots to avoid using a leading or a centralized device.
37
Figure 4.1: Architecture of the relationship among tasks