Mobile Robot Localization: a Modular, Odometry-Improving Approach

TL;DR

This work tackles robust localization for urban autonomous robots by addressing the fragility of relying on a single absolute pose source. It introduces a modular two-layer architecture that fuses outputs from off-the-shelf localization algorithms with raw sensor data via an EKF tailored to a vehicle-specific motion model, while online estimating model uncertainties to improve odometry during absolute-pose outages. The approach handles asynchronous and multi-rate measurements and can degrade gracefully when pose data are unavailable, demonstrated on a real TWIP robot (Yape) with GNSS and Cartographer, achieving substantial error reduction and resilience, including a reduction of cumulated position error from $5.3\mathrm{m}$ to $0.35\mathrm{m}$ over ~2 minutes without absolute measurements. The results indicate significant practical impact for autonomous last-mile systems, enabling continuous navigation in urban environments with reduced reliance on any single localization source.

Abstract

Despite the number of works published in recent years, vehicle localization remains an open, challenging problem. While map-based localization and SLAM algorithms are getting better and better, they remain a single point of failure in typical localization pipelines. This paper proposes a modular localization architecture that fuses sensor measurements with the outputs of off-the-shelf localization algorithms. The fusion filter estimates model uncertainties to improve odometry in case absolute pose measurements are lost entirely. The architecture is validated experimentally on a real robot navigating autonomously proving a reduction of the position error of more than 90% with respect to the odometrical estimate without uncertainty estimation in a two-minute navigation period without position measurements.

Figures (9)

• Figure 1: Block scheme of the proposed localization architecture.
• Figure 2: Yape, the deployment target of the developed localization algorithm.
• Figure 3: Sensitivity analysis of model uncertainties on the estimated (odometrical) trajectory: left wheel radius \ref{['fig:yape_arch/odometry_radius_sensitivity']} and IMU yaw rate bias \ref{['fig:yape_arch/odometry_imu_bias_sensitivity']}.
• Figure 4: Aerial view of the trajectory traveled by Yape in the courtyard of the Niguarda Hospital in Milan.
• Figure 5: Position (top) and yaw (bottom) error distribution of filter estimate with respect to Cartographer. The X axis refers to the fused position inputs (GNSS, Cartographer or both).