Mind the ground: A Power Spectral Density-based estimator for all-terrain rovers

TL;DR

This paper introduces a PSD-based method for real-time terrain unevenness estimation using onboard sensors, aiding autonomous navigation on uncharted natural terrains.

Contribution

It presents a novel PSD-based estimator for terrain roughness that is robust to vehicle tilt and applicable in real-time during normal rover operations.

Findings

Effective terrain roughness classification demonstrated in field tests.

Estimator shows limited sensitivity to vehicle tilt rotations.

Potential for integration into driving assistance systems.

Abstract

There is a growing interest in new sensing technologies and processing algorithms to increase the level of driving automation towards self-driving vehicles. The challenge for autonomy is especially difficult for the negotiation of uncharted scenarios, including natural terrain. This paper proposes a method for terrain unevenness estimation that is based on the power spectral density (PSD) of the surface profile as measured by exteroceptive sensing, that is, by using a common onboard range sensor such as a stereoscopic camera. Using these components, the proposed estimator can evaluate terrain on-line during normal operations. PSD-based analysis provides insight not only on the magnitude of irregularities, but also on how these irregularities are distributed at various wavelengths. A feature vector can be defined to classify roughness that is proved a powerful statistical tool for the characterization of a given terrain fingerprint showing a limited sensitivity to vehicle tilt rotations. First, the theoretical foundations behind the PSD-based estimator are presented. Then, the system is validated in the field using an all-terrain rover that operates on various natural surfaces. It is shown its potential for automatic ground harshness estimation and, in general, for the development of driving assistance systems.