Place classification with a graph regularized deep neural network model

TL;DR

This paper introduces a novel end-to-end deep learning framework for place classification in robotics, utilizing multi-layer laser data and graph regularization to improve accuracy and feature learning.

Contribution

It presents a new multi-layer deep neural network approach with graph regularization and confidence-based fusion for enhanced place classification.

Findings

Effective place classification demonstrated on experimental data.

Graph regularization improves local consistency and accuracy.

Automatically learned features compete with handcrafted statistical features.

Abstract

Place classification is a fundamental ability that a robot should possess to carry out effective human-robot interactions. It is a nontrivial classification problem which has attracted many research. In recent years, there is a high exploitation of Artificial Intelligent algorithms in robotics applications. Inspired by the recent successes of deep learning methods, we propose an end-to-end learning approach for the place classification problem. With the deep architectures, this methodology automatically discovers features and contributes in general to higher classification accuracies. The pipeline of our approach is composed of three parts. Firstly, we construct multiple layers of laser range data to represent the environment information in different levels of granularity. Secondly, each layer of data is fed into a deep neural network model for classification, where a graph regularization is imposed to the deep architecture for keeping local consistency between adjacent samples. Finally, the predicted labels obtained from all the layers are fused based on confidence trees to maximize the overall confidence. Experimental results validate the effective- ness of our end-to-end place classification framework in which both the multi-layer structure and the graph regularization promote the classification performance. Furthermore, results show that the features automatically learned from the raw input range data can achieve competitive results to the features constructed based on statistical and geometrical information.