Kidnapped Radar: Topological Radar Localisation using Rotationally-Invariant Metric Learning

Abstract

This paper presents a system for robust, large-scale topological localisation using Frequency-Modulated Continuous-Wave (FMCW) scanning radar. We learn a metric space for embedding polar radar scans using CNN and NetVLAD architectures traditionally applied to the visual domain. However, we tailor the feature extraction for more suitability to the polar nature of radar scan formation using cylindrical convolutions, anti-aliasing blurring, and azimuth-wise max-pooling; all in order to bolster the rotational invariance. The enforced metric space is then used to encode a reference trajectory, serving as a map, which is queried for nearest neighbours (NNs) for recognition of places at run-time. We demonstrate the performance of our topological localisation system over the course of many repeat forays using the largest radar-focused mobile autonomy dataset released to date, totalling 280 km of urban driving, a small portion of which we also use to learn the weights of the modified architecture. As this work represents a novel application for FMCW radar, we analyse the utility of the proposed method via a comprehensive set of metrics which provide insight into the efficacy when used in a realistic system, showing improved performance over the root architecture even in the face of random rotational perturbation.

Probabilistic Visual Place Recognition for Hierarchical Localization

“…Our proposed approach treats the quality scores from a place recognition algorithm as noisy sensor measurements and localizes using the Bayesian state estimation framework. In principle, place recognition methods using alternative sensor modalities such as LiDAR can also be used with our proposed methods [28]- [31]. 1 https://github.com/mingu6/ProbFiltersVPR.git…”

Radar-to-Lidar: Heterogeneous Place Recognition via Joint Learning

“…For radar-based place recognition, Kim et al (2020) extended the lidar-based handcrafted representation (Kim and Kim, 2018) to radar data directly. In Sȃftescu et al (2020), NetVLAD (Arandjelovic et al, 2016) was used to achieve radar-to-radar (R2R) place recognition. Then, the researchers used sequential radar scans to improve the localization performance (Gadd et al, 2020).…”

LiDAR Lateral Localisation Despite Challenging Occlusion from Traffic

“…Using dense maps in LiDAR-based approaches, such as Laser Odometry (LO) [13], is usually very accurate but computationally expensive for running in realtime. There is burgeoning interest in radar-based techniques, including Radar Odometry (RO) [14,15,16] and localisation [17]. As of yet, these techniques do not offer a rich enough understanding of scene semantics to be employed for road boundary problems.…”