Event
RadGeoLoc: A Particle Filter Radar Geolocation Method
1 reports · 1 sources
Shaunak Kolhe et al. propose RadGeoLoc, which aligns ground-based rotating radar measurements with satellite overhead RGB imagery for autonomous localization when GNSS is unreliable. The method uses radar representations pretrained via masked reconstruction and aligns them with satellite image features; radar odometry drives a particle filter to propose candidate poses, and a learned observation model provides the likelihood. Evaluation covers 8 datasets and 31 routes, spanning urban and off-road conditions, using only ground radar and a coarse local prior; in addition, the same route was traversed repeatedly over 13 months to examine robustness under weather changes. The paper has been submitted to ICRA 2027.
Report timeline
- RadGeoLoc: A Particle Filtering-Based Radar Geolocalization Method
RadGeoLoc proposes a geolocalization method that aligns ground-based spinning radar measurements with satellite RGB imagery for autonomous localization when GNSS is unreliable. The method uses a radar representation pretrained via masked reconstruction, aligns it with satellite image features, proposes candidate poses through a particle filter driven by radar odometry, and derives likelihoods from a learned observation model. Evaluation covers 8 datasets and 31 distinct routes spanning urban and offroad conditions, and examines robustness under changing weather through repeated traversals of the same route over 13 months, using only ground radar and a coarse local prior.
arXiv cs.RO ·