Balancing Act: Trading Off Doppler Odometry and Map Registration for Efficient Lidar Localization

TL;DR

This work tackles the challenge of achieving accurate lidar-based localization while maintaining real-time performance. It introduces a lightweight Doppler odometry integrated into a topometric localization pipeline and compares it to a high-accuracy ICP-based approach, while also exploring the impact of infrequent map matching by varying the localization cadence. Across five real-world routes and more than $100$ km of data, localizing every $n=10$ frames achieves sub-centimeter translational error and sub-degree rotational error with notable compute savings, and a knee-point analysis demonstrates real-time feasibility for both odometry strategies. The findings support a practical, adaptable localization framework where odometry choice and cadence are tuned to operational constraints, enabling efficient AV navigation with preserved accuracy.

Abstract

Most autonomous vehicles rely on accurate and efficient localization, which is achieved by comparing live sensor data to a preexisting map, to navigate their environment. Balancing the accuracy of localization with computational efficiency remains a significant challenge, as high-accuracy methods often come with higher computational costs. In this paper, we present two ways of improving lidar localization efficiency and study their impact on performance. First, we integrate a lightweight Doppler-based odometry method into a topometric localization pipeline and compare its performance against an iterative closest point (ICP)-based method. We highlight the trade-offs between these approaches: the Doppler estimator offers faster, lightweight updates, while ICP provides higher accuracy at the cost of increased computational load. Second, by controlling the frequency of localization updates and leveraging odometry estimates between them, we demonstrate that accurate localization can be maintained while optimizing for computational efficiency using either odometry method. Our experimental results show that localizing every 10 lidar frames strikes a favourable balance, achieving a localization accuracy below 0.05 meters in translation and below 0.1 degrees in orientation while reducing computational effort by over 30% in an ICP-based pipeline. We quantify the trade-off of accuracy to computational effort using over 100 kilometers of real-world driving data in different on-road environments.

Figures (3)

• Figure 1: We explore the trade-off between computational efficiency and localization accuracy with infrequent map matching, analyzing how dead-reckoning between these localization attempts impacts overall performance. Top right is our data collection platform, Boreas, equipped with an Aeva Aeries II FMCW lidar sensor.
• Figure 2: The structure of the pose graph during submap construction and localization, comparing the localization pipeline from Burnett2022Ready to the modified pipeline used in this work.
• Figure 3: Pareto front plots for the longitudinal (top), lateral (middle), and yaw (bottom) RMSE, comparing Doppler and ICP-based odometry methods across three routes. The plots show the trade-off between error and computational effort, with markers representing each tested interval. For the Doppler pipeline, the intervals are $n = 1,2,5,10,15,25,50$, while for the ICP pipeline, the intervals are $n = 1,2,5,10,15,25,50,100, 150$.