Mobility-Aware Multi-Objective Task Allocation for Wireless Sensor Networks

Survey on Multi-Objective Task Allocation Algorithms for IoT Networks

“…Reliability ( R ) is defined as the time of the first fault leading to the network not being able to execute the global task T divided by the Network Lifetime [ 11 ]. …”

“…Weikert et al [ 9 , 11 , 24 , 25 ] proposed a group of EAs suited to deal with the challenges of task allocation in mobile and failure-prone networks. Their latest work [ 9 ] combines these approaches and improves upon the weaknesses of the earlier versions.…”

“…With a few exceptions, the RL approaches were only applied to the static problem, and no approach incorporated a multi-objective design. However, RL and especially DRL approaches may be useful as part of other algorithms, such as surrogate models used for reducing the computational complexity [ 56 ] or the prediction of node dynamics, as shown by Weikert et al [ 11 ]. GNNs especially are a promising approach to develop new solutions or enhance existing ones for the optimization problems of IoT networks.…”

“…Five of the reviewed works [ 9 , 11 , 47 , 48 , 50 ] incorporate a mechanism for preemptively planning the task allocations around the movement of nodes. However, two of these works assume either full knowledge of the trajectories [ 50 ] or actively plan the trajectories to improve the task allocation optimization [ 48 ].…”

“…To deal with the unknown future positions of nodes, only given the current node states, the remaining works all use a DRL-based approach in their optimization. This is either achieved by predicting future node positions [ 9 , 11 ] that are then used in the optimization process, or via directly optimizing based on the predicted reward of the network [ 47 ]. Notably, the works relying on predicted nodes can use any prediction mechanism in their algorithm and are not necessarily tied to a DRL approach.…”