An Effective Method for Identifying Clusters of Robot Strengths

TL;DR

This paper introduces a new method for identifying clusters of robot strengths in competitive robotics data, improving model accuracy by reducing overparameterization and allowing flexible cluster reassignments.

Contribution

A novel combined hierarchical and non-hierarchical classification approach with model selection criteria for effective robot strength clustering.

Findings

Accurate estimation of robot clusters and strengths.

Reduced overestimation of clusters using BIC.

Enhanced model selection with nested and monotonic indices.

Abstract

In the analysis of qualification data from the FIRST Robotics Competition, the ratio of the number of observations to the number of parameters has been found to be quite small for the commonly used winning margin power rating (WMPR) model. This usually leads to imprecise estimates and inaccurate predictions in such a three-on-three game. With the finding of a clustering feature in estimated robot strengths, a more flexible model with latent clusters of robots was proposed to alleviate overparameterization of the WMPR model. Since its structure can be regarded as a dimension reduction of the parameter space in the WMPR model, the identification of clusters of robot strengths is naturally transformed into a model selection problem. Instead of comparing a huge number of competing models, we develop an effective method to estimate the number of clusters, clusters of robots, and robot strengths. The new method consists of two parts: (i) a combination of hierarchical and non-hierarchical classifications to determine candidate models; and (ii) variant goodness-of-fit criteria to select optimal models. Different from existing hierarchical classification systems, each step of ours is based on estimated robot strengths from a candidate model in the preceding non-hierarchical classification step. A great advantage of the designed non-hierarchical classification system is to examine the possibility of reassigning robots to other cluster sets of robots. To reduce the overestimation of clusters by the mean squared prediction error criteria, the corresponding BIC are established as alternatives for model selection. By assembling these essential elements into a coherent whole, a systematic procedure is presented to perform the estimation. In addition, we propose two indices to measure the nested relation between cluster sets of two models and monotonic association between robot strengths of two models.