WRAP: Wasserstein-Robust Adaptive Plug-in for Robot Localization

2026-08-10

AI Summary

This work addresses the susceptibility of robot localization to bias and covariance miscalibration under dynamic perceptual conditions. The authors propose WRAP, an adapter-agnostic robust plugin that, for the first time, integrates Wasserstein robust optimization into a nonlinear EKF/ESKF framework. WRAP employs a causal module to estimate time-varying statistics and introduces a mean-preserving Wasserstein local update that decouples mean adaptation from covariance robustification, utilizing distinct uncertainty radii for propagation and perception. Experiments demonstrate that WRAP reduces 3D position RMSE by 27.4% over the nominal ESKF across 18 unseen UWB–IMU sequences. GNSS–INS evaluations confirm that mean adaptation primarily drives accuracy gains, while directional redistribution enhances consistency. A single robust update requires only 0.05 ms (UWB) to 2.92 ms (GNSS) on a Jetson Orin Nano.

Abstract

Robotic localization under changing sensing conditions can suffer from biased errors and miscalibrated covariances. We present WRAP, an adapter-agnostic Wasserstein-robust plug-in for nonlinear extended Kalman filter (EKF) and error-state Kalman filter (ESKF) stacks. A causal module supplies time-varying effective process and measurement statistics; a mean-preserving Wasserstein local update then computes least-favorable covariances and a robust gain without changing the propagation model, residual, or retraction. This separates mean adaptation from covariance robustification and uses distinct radii for propagation and sensing. On 18 UWB--IMU sequences held out from adapter training, adapter-only and WRAP reduce mean 3-D position RMSE by $19.8\%$ and $27.4\%$ relative to the nominal ESKF; an isotropic ablation reaches $19.5\%$, linking the incremental gain to directional process-covariance redistribution. An in-sample GNSS--INS study shows that mean adaptation provides most of the accuracy gain, while DR improves consistency and mitigates over-tightened classical covariance estimates. The robust solve takes 0.05 ms for UWB and 2.92 ms for GNSS on a Jetson Orin Nano.

Problem

• robot localization
• biased errors
• miscalibrated covariances
• changing sensing conditions
• Wasserstein robustness

Innovation

• Wasserstein robustness
• adaptive Kalman filtering
• covariance robustification
• robot localization
• uncertainty quantification