Hallucinating robots: Inferring Obstacle Distances from Partial Laser Measurements

TL;DR

This paper introduces a neural network-based method to estimate obstacle distances from partial 2D laser scans, enabling robots to detect obstacles like glass panels that are invisible to standard sensors.

Contribution

The work presents a novel neural network autoencoder approach to infer obstacle distances directly from raw 2D laser data, improving perception without additional sensors.

Findings

The proposed model accurately predicts obstacle distances from partial laser data.

Qualitative tests on a Care-O-bot 4 demonstrate real-time obstacle inference.

The method outperforms baseline approaches in obstacle detection accuracy.

Abstract

Many mobile robots rely on 2D laser scanners for localization, mapping, and navigation. However, those sensors are unable to correctly provide distance to obstacles such as glass panels and tables whose actual occupancy is invisible at the height the sensor is measuring. In this work, instead of estimating the distance to obstacles from richer sensor readings such as 3D lasers or RGBD sensors, we present a method to estimate the distance directly from raw 2D laser data. To learn a mapping from raw 2D laser distances to obstacle distances we frame the problem as a learning task and train a neural network formed as an autoencoder. A novel configuration of network hyperparameters is proposed for the task at hand and is quantitatively validated on a test set. Finally, we qualitatively demonstrate in real time on a Care-O-bot 4 that the trained network can successfully infer obstacle distances from partial 2D laser readings.