Using Convolutional Neural Networks in Robots with Limited Computational Resources: Detecting NAO Robots while Playing Soccer

TL;DR

This paper explores how to effectively implement CNNs on resource-limited robots, proposing two real-time detectors for NAO robots in soccer, achieving high detection accuracy with minimal processing time.

Contribution

It introduces general design guidelines for CNN deployment on limited-resource robots and presents two novel real-time detectors based on XNOR-Net and SqueezeNet.

Findings

Detectors process each proposal in ~1ms

Achieved detection rate of ~97%

Able to run in real-time during soccer games

Abstract

The main goal of this paper is to analyze the general problem of using Convolutional Neural Networks (CNNs) in robots with limited computational capabilities, and to propose general design guidelines for their use. In addition, two different CNN based NAO robot detectors that are able to run in real-time while playing soccer are proposed. One of the detectors is based on the XNOR-Net and the other on the SqueezeNet. Each detector is able to process a robot object-proposal in ~1ms, with an average number of 1.5 proposals per frame obtained by the upper camera of the NAO. The obtained detection rate is ~97%.