Set-membership information fusion for multisensor nonlinear dynamic systems

Zhiguo Wang, Xiaojing Shen, Yunmin Zhu

TL;DR

This paper tackles the challenging problem of fusing data from multiple sensors in nonlinear dynamic systems, where noise is unknown but bounded. The authors introduce novel analytical fusion formulas for both centralized and distributed architectures, which are surprisingly similar to classic information filters and avoid the need to know sensor noise correlations. They also propose a multi-algorithm fusion approach that further improves accuracy by combining multiple estimation strategies.

What Problem Does This Paper Address?

Estimating the state of a dynamic system (like tracking a moving object) using data from multiple sensors is a fundamental problem in many fields. When the system is nonlinear and the noise affecting the sensors is not precisely known but can be bounded (e.g., "the error is at most X"), this becomes significantly more complex than standard linear estimation. Traditional methods often rely on knowing the exact statistical properties of noise, which is not always feasible. This paper focuses on "set-membership" estimation, where the goal is to find a bounding region (an ellipsoid) that is guaranteed to contain the true state, even with unknown but bounded noise. The challenge is to efficiently fuse these bounding ellipsoids from multiple sensors, especially in nonlinear systems, where analytical solutions are hard to come by.

Methods

The paper proposes two main fusion architectures:

  • Centralized Fusion: All sensor data is sent to a central fusion center.
  • Distributed Fusion: Each sensor processes its data locally, and then these local estimates are fused.

For both architectures, the authors achieve the following:

  1. Semidefinite Programming (SDP) Formulation: They convert the fusion problem into an SDP, a type of optimization problem that can be efficiently solved.
  2. Analytical Fusion Formulas: Using a "decoupling technique," they derive explicit, analytical formulas for the fusion process. These formulas are notable for their similarity to the well-established information filter, a common tool in linear estimation.
  3. No Correlation Knowledge Needed: A key advantage is that these analytical formulas do not require knowledge of the correlations between the measurement noises of different sensors.
  4. Multi-Algorithm Fusion: To further refine the state estimate, they introduce a multi-algorithm fusion strategy. This involves using multiple parallel algorithms with different weighting objectives to generate several bounding ellipsoids, and then taking their intersection to obtain a tighter, more accurate bounding ellipsoid.

Datasets

Not discussed in the paper. The paper focuses on theoretical development and uses a numerical example for evaluation.

Key Insights and Ideas

  • Analytical Solutions for Nonlinear Set-Membership Fusion: Deriving analytical formulas for set-membership fusion in nonlinear systems is a significant challenge due to the inherent nonconvexity. The paper's success in finding such formulas is a major breakthrough.
  • Information Filter Analogy: The derived analytical formulas closely resemble the structure of the classic information filter. This suggests a unifying framework for estimation under different noise assumptions (probabilistic vs. bounded).
  • Decoupling Technique's Power: The use of a decoupling technique proves crucial in simplifying the complex nonlinear fusion problem into manageable analytical forms.
  • Sensor Correlation Independence: The ability to fuse sensor data without knowing cross-sensor noise correlations is a practical advantage, as such correlations are often difficult to estimate accurately.
  • Multi-Algorithm Fusion for Tighter Bounds: Combining multiple estimation strategies, each optimized for different aspects of the state, leads to a more precise bounding ellipsoid, demonstrating the benefit of complementary approaches.

Key Results

The paper presents theoretical derivations for centralized and distributed set-membership information fusion algorithms. A numerical example in target tracking demonstrates the effectiveness of these algorithms. The results show that:

  • Both centralized and distributed fusion perform better than individual sensors.
  • The multi-algorithm fusion approach consistently yields tighter error bounds (smaller bounding ellipsoids) compared to both centralized and distributed fusion, especially in nonlinear dynamic systems. This is attributed to its ability to leverage information from each state component through weighted objectives.

Limitations and Future Work

  • Computational Complexity: While the use of SDP and decoupling techniques improves efficiency, the computational complexity can still be significant for high-dimensional systems. The paper notes complexity is roughly $O(n^3)$ for centralized fusion and $O(n^3L + m^3L^4)$ for distributed fusion, where $n$ is state dimension, $m$ is measurement dimension, and $L$ is the number of sensors.
  • Future Work: The authors suggest future research directions including:
  • Extending these methods to multisensor nonlinear dynamic systems with multiple targets.
  • Investigating sensor management strategies.
  • Developing fusion techniques for heterogeneous sensors.

Summary updated

Abstract

The set-membership information fusion problem is investigated for general multisensor nonlinear dynamic systems. Compared with linear dynamic systems and point estimation fusion in mean squared error sense, it is a more challenging nonconvex optimization problem. Usually, to solve this problem, people try to find an efficient or heuristic fusion algorithm. It is no doubt that an analytical fusion formula should be much significant for raising accuracy and reducing computational burden. However, since it is a more complicated than the convex quadratic optimization problem for linear point estimation fusion, it is not easy to get the analytical fusion formula. In order to overcome the difficulty of this problem, two popular fusion architectures are considered: centralized and distributed set-membership information fusion. Firstly, both of them can be converted into a semidefinite programming problem which can be efficiently computed, respectively. Secondly, their analytical solutions can be derived surprisingly by using decoupling technique. It is very interesting that they are quite similar in form to the classic information filter. In the two analytical fusion formulae, the information of each sensor can be clearly characterized, and the knowledge of the correlation among measurement noises across sensors are not required. Finally, multi-algorithm fusion is used to minimize the size of the state bounding ellipsoid by complementary advantages of multiple parallel algorithms. A typical numerical example in target tracking demonstrates the effectiveness of the centralized, distributed, and multi-algorithm set-membership fusion algorithms. In particular, it shows that multi-algorithm fusion performs better than the centralized and distributed fusion.
BibTeX
@misc{wang2017set,
  author = {Zhiguo Wang and Xiaojing Shen and Yunmin Zhu},
  title = {{Set-membership information fusion for multisensor nonlinear dynamic systems}},
  year = {2017},
  eprint = {1702.05214},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1702.05214},
}
Figure