Systematic Error Modeling and Bias Estimation

Feihu Zhang, Alois C. Knoll

TL;DR

This paper tackles the challenge of systematic errors in cooperative localization systems, which arise when sensors transform data into a common reference frame. The authors propose a novel method to model these errors statistically and then estimate sensor biases using a weighted nonlinear least squares approach, demonstrating improved performance over existing methods.

What Problem Does This Paper Address?

Cooperative localization (CL) systems rely on multiple sensors to gather data and improve the accuracy and reliability of localization. A crucial step in CL is transforming data from individual sensors into a common reference system. This transformation process, however, often introduces systematic errors, which are consistent deviations from the true value, distinct from random noise. These systematic errors can significantly degrade the overall localization accuracy. Existing methods for sensor registration and bias estimation often struggle with target maneuvers or require ground truth information, which is not always available in real-world scenarios. This paper aims to address this gap by developing a method to model and estimate these systematic errors and sensor biases robustly, even in the presence of maneuvers and without relying on ground truth.

Methods

The core of the paper's contribution lies in two main areas:

  • Systematic Error Modeling: The authors first analyze the statistical properties of systematic errors in terms of range and bearing. They derive mathematical models for the expectation and covariance of these errors, considering the effects of biases and noise. Crucially, they then adapt these models to be conditioned on the actual sensor measurements, making them practical for real-world application.
  • Bias Estimation: Building upon the error models, a weighted nonlinear least squares (NLS) method is employed to estimate the sensor biases (specifically, range bias $r_b$ and bearing bias $\theta_b$). This method utilizes information from an additional sensor (like GPS, which is assumed to have only random noise) to provide a reference. The weights in the NLS are derived from the estimated covariances of the systematic errors, allowing the method to adapt to varying error characteristics. This approach is designed to handle target maneuvers by normalizing the weights of pseudo-measurements.

Datasets

The paper does not introduce or use specific named datasets. Instead, it relies on simulated data for evaluation. The simulations involve three objects with different motion patterns: stationary, constant velocity, and maneuvering (constant turn). These simulations are run for 50 steps, with 1000 Monte-Carlo trials to assess the statistical properties and performance of the proposed method.

Key Insights and Ideas

  • Systematic errors are predictable and modelable: The paper demonstrates that systematic errors, while challenging, possess statistical properties that can be modeled. This opens the door for their explicit estimation and compensation.
  • Measurement-conditioned error models are key: Moving from theoretical error models to those conditioned on actual measurements is a critical step for practical application in real-time systems.
  • Weighted NLS with adaptive weights handles maneuvers: The use of weighted NLS, where weights are derived from estimated error covariances, allows the bias estimation to be robust to target maneuvers, a significant improvement over methods that struggle with dynamic targets.
  • Performance improves with proximity to radar: The study observes that both the estimated expectation and covariance of systematic errors decrease significantly when objects are closer to the radar, suggesting that sensor measurements are more reliable in such scenarios.

Key Results

The paper presents simulation results that highlight the effectiveness of the proposed approach:

  • Accurate Bias Estimation: The proposed method successfully estimates biases in range and bearing, converging towards the ground truth values over time, even for maneuvering objects.
  • Superior Performance: Quantitative evaluation using Root Mean Square Error (RMSE) shows that the proposed approach consistently outperforms a "Normal Least Square" method (likely a standard linear least squares approach with linearization) in both range and bearing bias estimation. The RMSE for the proposed method is generally lower, indicating more accurate and stable bias estimates.
  • Robustness to Maneuvers: The simulations with maneuvering objects demonstrate that the proposed method maintains good performance, unlike some traditional methods that might falter in dynamic scenarios.

Limitations and Future Work

  • Reliance on Additional Sensor: The bias estimation method requires information from an additional sensor (e.g., GPS) to provide a reference for the transformation. The performance might be affected by the quality and noise characteristics of this auxiliary sensor.
  • Computational Complexity: While not explicitly detailed, nonlinear least squares methods can be computationally intensive, which might be a consideration for real-time applications with limited processing power.
  • Future Work: The authors suggest that future work will focus on applying the proposed approach in real-world scenarios. This would involve testing the method on actual sensor data and evaluating its performance in complex, unsimulated environments. They also mention that the unbiased measurement can be calculated during the filtering phase, combined with uncertainties, which could be a direction for further development.

Summary updated

Abstract

This paper analyzes the statistic properties of the systematic error in terms of range and bearing during the transformation process. Furthermore, we rely on a weighted nonlinear least square method to calculate the biases based on the proposed models. The results show the high performance of the proposed approach for error modeling and bias estimation.
BibTeX
@inproceedings{zhang2016systematic,
  author = {Feihu Zhang and Alois C. Knoll},
  title = {{Systematic Error Modeling and Bias Estimation}},
  booktitle = {Sensors},
  year = {2016},
  url = {https://openreview.net/forum?id=ZonE1_Og7oY},
}

Loading related content...

Figure