Safe Robot Navigation Via Multi-Modal Anomaly Detection

Abstract

Navigation in natural outdoor environments requires a robust and reliable traversability classification method to handle the plethora of situations a robot can encounter. Binary classification algorithms perform well in their native domain but tend to provide overconfident predictions when presented with out-of-distribution samples, which can lead to catastrophic failure when navigating unknown environments. We propose to overcome this issue by using anomaly detection on multi-modal images for traversability classification, which is easily scalable by training in a self-supervised fashion from robot experience. In this work, we evaluate multiple anomaly detection methods with a combination of uni-and multi-modal images in their performance on data from different environmental conditions. Our results show that an approach using a feature extractor and normalizing flow with an input of RGB, depth and surface normals performs best. It achieves over 95% area under the ROC curve and is robust to out-of-distribution samples.

A Unifying Review of Deep and Shallow Anomaly Detection

“…An advantage of these models over other methods is that one can calculate the likelihood of a point directly without any approximation while also being able to sample from it reasonably efficiently. Because the density px(x) can be computed exactly, normalizing flow models can be applied directly for AD [288], [289].…”

Rough Terrain Navigation for Legged Robots using Reachability Planning and Template Learning

“…While we have argued in previous work [5], [6] that purely geometric approaches are not sufficient for navigation in natural outdoor environments, approaches relying on semantic information exhibit the same issues as traditional geometric approaches. They, either implicitly through semantic segmentation of the environment [12], [13], [14], [15], 1 www.github.com/leggedrobotics/art_planner or explicitly [16], [17], [18] predict a traversability label.…”

“…As we have shown in previous work [5], [6], we can learn to predict non-geometric obstacles like slippery or unknown terrain from visual semantic information. We can regard this information as foothold feasibility classification and only allow safe regions to support footholds by removing unsafe terrain from the reachability collision map.…”

An Online Robot Collision Detection and Identification Scheme by Supervised Learning and Bayesian Decision Theory

“…Secondly, an accidental collision is usually referred to as an instantaneous anomaly [14], which occurs unexpectedly and only lasts for a short period of time. Thus, it is more difficult to be identified than the other anomalies that are featured with low bandwidth and steady changes [15], [16]. The main target of this paper is to fill these gaps by proposing a novel online CDI scheme that ensures both high accuracy and fast response.…”