Learning Optimal Topology for Ad-hoc Robot Networks

TL;DR

This work tackles the challenge of predicting optimal topology in ad-hoc robot networks, where the topology must balance connectivity, reliability, and congestion. The authors transform a complex multi-task topology problem into multiple per-robot multi-class classification tasks and synthesize OpTopNET, a stacked-ensemble predictor that combines three low-level classifiers per robot with an XGBoost blender. Ground-truth topologies are generated via a formal definition involving a backbone cycle and branches, parameterized by a connectivity threshold $\delta$ and tension bound $\epsilon$, with backbone computation known to be $NP$-hard. Experiments on a 10-robot network show the ensemble achieving about 81% average test accuracy, outperforming a graph-normalized CNN baseline, demonstrating a practical, data-driven path to real-time topology guidance in dynamic robotic networks.

Abstract

In this paper, we synthesize a data-driven method to predict the optimal topology of an ad-hoc robot network. This problem is technically a multi-task classification problem. However, we divide it into a class of multi-class classification problems that can be more efficiently solved. For this purpose, we first compose an algorithm to create ground-truth optimal topologies associated with various configurations of a robot network. This algorithm incorporates a complex collection of optimality criteria that our learning model successfully manages to learn. This model is an stacked ensemble whose output is the topology prediction for a particular robot. Each stacked ensemble instance constitutes three low-level estimators whose outputs will be aggregated by a high-level boosting blender. Applying our model to a network of 10 robots displays over 80% accuracy in the prediction of optimal topologies corresponding to various configurations of the cited network.

Figures (4)

• Figure 1: Interrelated correlations among robot coordinates in view of each other's classifiability. (The axes of each subfigure are absolute correlation factors. For example, the scattering dependency of the classifiability of the horizontal coordinate component of robot 5, i.e., X5, is displayed with respect to the coordinate components of robots 1 and 2. One can observe that if the coordinates of a particular robot, say, X1-Y1, are taken into account, the separation boundary to classify X5 is different than the correlated coordinate cases, such as X1-Y2 or X2-Y1. In these cases, the classifiability of X5 exhibits more convoluted trends.)
• Figure 2: Learning model specification
• Figure 3: The average classification accuracy of ensemble classifiers and those of their constituent classifiers. (The vertical dashed line represents the overall average accuracy of the ensembles.)
• Figure 4: Average accuracy and F1 scores associated with OpTopNET and GnCN

Theorems & Definitions (14)

• Definition 2: Robot Network
• Definition 3: Reliable Set
• Definition 4: Critical Set
• Definition 5: Link Set
• Definition 6: Robot Network Graph
• Definition 7: Backbone Cycle
• Definition 8: Indirect Reachability
• Definition 9: Branch
• Definition 10: Robot Network Topology
• Remark 11