Mobile Sensor Path Planning for Kalman Filter Spatiotemporal Estimation

Jiazhong Mei, Steven L. Brunton, J. Nathan KutzSensors 2024 · DOI

TL;DR

This paper tackles the challenge of accurately estimating spatiotemporal data using limited mobile sensors. They propose a novel greedy path planning algorithm that optimizes sensor trajectories to maximize the observability of the system, leading to improved Kalman filter estimation performance and faster convergence, even outperforming multiple stationary sensors.

What Problem Does This Paper Address?

Estimating spatiotemporal data (data that varies in both space and time) is crucial across many scientific fields. Traditionally, this is done with stationary sensors. However, limited stationary sensors can struggle to capture complex dynamics, especially those with localized structures. While mobile sensors offer a potential solution, efficiently planning their paths to maximize estimation accuracy is a significant challenge. Existing methods often focus on stationary sensors or have limitations in scalability and computational efficiency. This paper addresses the problem of optimally planning the trajectories of a limited number of mobile sensors to achieve accurate spatiotemporal estimation using Kalman filtering, particularly for systems with complex dynamics.

Methods

The core of the paper lies in developing a greedy path planning algorithm for mobile sensors. The approach is grounded in the concept of observability, which dictates how well the system's internal states can be inferred from its measurements.

  1. Low-Rank Representation: To handle high-dimensional spatiotemporal data efficiently, the authors first employ Dynamic Mode Decomposition (DMD) to find a low-rank representation of the system's dynamics. This reduces the computational complexity without significant loss of information.
  2. Observability Metric: The key metric for optimizing sensor paths is the condition number of the projected observability matrix. A lower condition number indicates better observability and, consequently, better Kalman filter estimation performance.
  3. Greedy Path Planning: A time-forwarding greedy algorithm is proposed. At each step, it selects the next sensor location that minimizes the condition number of the observability matrix, considering constraints like sensor speed and the need to form a periodic trajectory. This algorithm leverages techniques similar to QRcp and GappyPOD for efficient submatrix selection.
  4. Kalman Filtering: The planned sensor trajectories are then used within a standard Kalman filter framework to estimate the spatiotemporal data. The paper analyzes how three key timescales – the system's dynamics, sensor velocity, and sampling rate – influence estimation performance.

Datasets

The paper uses two primary examples for evaluation:

  1. Synthetic Data: A sparse linear dynamical system on a torus is generated to demonstrate the algorithm's effectiveness in capturing both global and local structures.
  2. Sea Surface Temperature (SST) Dataset: Weekly optimum interpolated SST data from NOAA is used. This dataset is characterized by global features, allowing for the evaluation of Kalman filter convergence and the benefits of mobile sensing in such scenarios.

Key Insights and Ideas

  • Mobile Sensors as a Game Changer: Mobile sensors, when strategically planned, can achieve estimation performance comparable to or even exceeding that of a significantly larger number of stationary sensors.
  • Observability is Key: The condition number of the observability matrix is a direct indicator of Kalman filter estimation quality. Optimizing sensor paths to improve this condition number is a powerful strategy.
  • Timescale Matters: The interplay between the system's intrinsic dynamics, sensor speed, and sampling rate critically influences how well a mobile sensor can observe and estimate the system.
  • Greedy Approach Works: A greedy, time-forwarding approach to path planning, focused on immediate observability gains, proves effective and computationally scalable.
  • Local Structures Benefit Most: Mobile sensors are particularly advantageous for systems with spatially localized structures, as they can actively move to capture these features, which stationary sensors might miss.

Key Results

  • In the synthetic example, one mobile sensor with an optimized path achieved performance equivalent to three stationary sensors for capturing local Gaussian modes.
  • For the Kuramoto-Sivashinsky (KS) system, a chaotic and nonlinear system, mobile sensors significantly improved estimation accuracy compared to fixed sensors, with performance approaching that of 100 fixed sensors when sensor speed was high.
  • On the Sea Surface Temperature (SST) dataset, one mobile sensor demonstrated much faster convergence of the Kalman filter estimation error compared to stationary sensors, with higher speeds leading to quicker convergence.
  • The proposed greedy path planning algorithm is shown to be scalable and computationally efficient, outperforming previous methods.

Limitations and Future Work

  • Sensor Speed Constraint: The current algorithm treats sensor speed as a hard constraint. The authors suggest future work could incorporate speed as a cost to be minimized within a weighted objective function.
  • Energy Efficiency: The paper acknowledges that energy consumption is not explicitly considered. Future work could include planning energy-efficient trajectories.
  • Background Flow: For applications like flow field estimation, the influence of background flow on sensor movement is neglected. Incorporating this into the path planning is a promising extension.
  • Asynchronous Trajectories: The current work focuses on synchronized periodic trajectories. Exploring asynchronous trajectories for individual sensors could be beneficial for multiscale systems.
  • Nonlinear Systems: While the paper uses DMD for linear approximation, exploring better data-driven linear or nonlinear system identification methods could further improve performance.

Summary updated

Abstract

The estimation of spatiotemporal data from limited sensor measurements is a required task across many scientific disciplines. In this paper, we consider the use of mobile sensors for estimating spatiotemporal data via Kalman filtering. The sensor selection problem, which aims to optimize the placement of sensors, leverages innovations in greedy algorithms and low-rank subspace projection to provide model-free, data-driven estimates. Alternatively, Kalman filter estimation balances model-based information and sparsely observed measurements to collectively make better estimation with limited sensors. It is especially important with mobile sensors to utilize historical measurements. We show that mobile sensing along dynamic trajectories can achieve the equivalent performance of a larger number of stationary sensors, with performance gains related to three distinct timescales: (i) the timescale of the spatiotemporal dynamics, (ii) the velocity of the sensors, and (iii) the rate of sampling. Taken together, these timescales strongly influence how well-conditioned the estimation task is. We draw connections between the Kalman filter performance and the observability of the state space model and propose a greedy path planning algorithm based on minimizing the condition number of the observability matrix. This approach has better scalability and computational efficiency compared to previous works. Through a series of examples of increasing complexity, we show that mobile sensing along our paths improves Kalman filter performance in terms of better limiting estimation and faster convergence. Moreover, it is particularly effective for spatiotemporal data that contain spatially localized structures, whose features are captured along dynamic trajectories.
BibTeX
@inproceedings{mei2024mobile,
  author = {Jiazhong Mei and Steven L. Brunton and J. Nathan Kutz},
  title = {{Mobile Sensor Path Planning for Kalman Filter Spatiotemporal Estimation}},
  booktitle = {Sensors},
  year = {2024},
  doi = {10.3390/s24123727},
}

Loading related content...

Figure