Development of Relative Positioning Technique for Swarm Robots Using Distance Information Based on Deep Learning in Indoor Environments

This paper proposes two DNNs for predicting the relative positions of mobile swarm nodes in real-time, utilizing only distance information.The first method simultaneously estimates the coordinates of each node using acquired distance information.The second method estimates the coordinates by grouping a total of four nodes, including three anchor nodes.To resolve the ambiguity of the estimated coordinates, constraints are applied to the nodes playing the role of anchor nodes.To compare the performance of the DNN, a grid-based algorithm is adopted as the conventional method.Limiting the number of nodes on the coordinate plane to 4 to 8 in simulated experiments, the results indicate a degradation in coordinate estimation performance as the standard deviation (SD) of noise increases, across all methods.Additionally, except for scenarios involving 8 nodes with an SD of 0.02 m, the proposed technique exhibits superior performance in all cases.