Selective memory: Recalling relevant experience for long‐term visual localization

Abstract

Visual navigation is a key enabling technology for autonomous mobile vehicles. The ability to provide large-scale, long-term navigation using low-cost, low-power vision sensors is appealing for industrial applications. A crucial requirement for long-term navigation systems is the ability to localize in environments whose appearance is constantly changing over time-due to lighting, weather, seasons, and physical changes. This paper presents a multiexperience localization (MEL) system that uses a powerful map representation-storing every visual experience in layers-that does not make assumptions about underlying appearance modalities and generators. Our localization system provides real-time performance by selecting online, a subset of experiences against which to localize. We achieve this task through a novel experience-triage algorithm based on collaborative filtering, which selects experiences relevant to the live view, outperforming competing techniques. Based on classical memory-based recommender systems, this technique also enables landmarklevel recommendations, is entirely online, and requires no training data. We demonstrate the capabilities of the MEL system in the context of long-term autonomous path following in unstructured outdoor environments with a challenging 100-day field experiment through day, night, snow, spring, and summer. We furthermore provide offline analysis comparing our system to several state-of-theart alternatives. We show that the combination of the novel methods presented in this paper enable full use of incredibly rich multiexperience maps, opening the door to robust long-term visual localization.

A Bayesian tracker for synthesizing mobile robot behaviour from demonstration

“…The path is split into local maps and a key question is to choose the relevant one into which localising the robot. Recent approaches use recommendation techniques (collaborative filtering) for this selection, as MacTavish et al (2018), or implicit encoding through a neural network, as Xie et al (2020). However, they use this information to solve the geometric localisation problem before applying standard trajectory planning and following techniques.…”

Accurate outdoor ground truth based on total stations

“…Regarding the robotic platforms, there are almost no limitations; the only one being how many prisms the robot can carry. Researchers used total stations to track skid steered robots [9], [10], a tethered wheeled robot [11], planetary rover [12], [13], unmanned surface vessel [14], and walking robots [4], [15]. In our application, we use a large skid-steered robot suitable for winter in subarctic forest.…”

Long-term map maintenance pipeline for autonomous vehicles

“…In [7] authors describe a graph-based SLAM [8] approach which was designed to allow merging of new map fragments to the localisation map. MacTavish et al in [9] present a methodology to recall relevant experiences in visual localisation, the multi-experience localisation system stores every visual experience in a layer and selects the one more suitable for the current view.…”