TL;DR
This paper presents a compact, low-cost embedded system for micro underwater vehicles (μAUVs) to determine their position using the attenuation of electromagnetic (EM) waves. This approach overcomes the limitations of traditional localization methods in confined underwater environments and is crucial for enabling autonomous μAUV operations.
What Problem Does This Paper Address?
Deploying autonomous underwater vehicles (AUVs), especially micro-sized ones (μAUVs), in confined underwater environments like tanks presents a significant challenge for self-localization. Traditional methods like GPS are unusable underwater due to signal attenuation. Camera-based systems struggle with poor visibility and require off-board processing, leading to latency and bandwidth limitations, especially for large fleets. Existing underwater localization methods often rely on acoustic signals, which are susceptible to multipath propagation, reflections, and reverberation, and require precise clock synchronization, a difficult feat for μAUVs. This paper addresses the need for a reliable, low-cost, and embedded self-localization system for μAUVs in such challenging environments.
Methods
The core of the proposed system is an adaptation of a previously developed EM wave attenuation-based localization method to a μAUV-compatible architecture. This method relies on the principle of spherical localization, where the receiver estimates its distance to multiple beacons based on the received signal strength (RSS).
Key methodological aspects include:
- Underwater Range Sensor Model (URSM): This model, adapted from prior work, relates the received signal strength (RSS) to the distance between the receiver and a beacon. It accounts for the rapid attenuation of EM waves in water.
- Signal Identification Using Channel Allocation: To distinguish signals from multiple beacons transmitting simultaneously, each beacon emits its signal at a unique frequency. A Fast Fourier Transform (FFT) is used by the receiver to separate these signals and determine their respective RSS values.
- Embedded Hardware Architecture: The system utilizes off-the-shelf, low-cost components. A modified DVB-T USB dongle acts as a software-defined radio (SDR) to capture EM spectrum segments and compute power density spectra. A single-board computer (SBC), specifically a Raspberry Pi Zero, handles the signal processing, localization algorithms, and interfacing.
- Localization Algorithms: Both an Extended Kalman Filter (EKF) and a Particle Filter (PF) are implemented to fuse the estimated ranges from the beacons and compute the receiver's position. The PF is noted to potentially outperform the EKF in scenarios with strong nonlinearities and large noise variance.
- Calibration: A crucial step involves calibrating the URSM parameters for each beacon by measuring RSS at various known positions and fitting the model using a non-linear least-squares algorithm.
Datasets
The paper describes experiments conducted in a water tank at the Institute of Mechanics and Ocean Engineering, Hamburg University of Technology. The authors collected RSS measurements on a grid within this tank. These measurements were used to:
- Validate the URSM by fitting its parameters.
- Analyze the EM field characteristics of the test tank.
- Evaluate the performance of the static and dynamic localization algorithms.
No pre-existing public datasets were used; the data was generated from the experimental setup described.
Key Insights and Ideas
- EM attenuation is a viable localization cue underwater: Despite the challenges of water, the rapid attenuation of EM waves provides a measurable signal characteristic that can be reliably used for range estimation, unlike acoustic signals which suffer more from reflections.
- Low-cost embedded systems can achieve μAUV localization: The successful integration of a DVB-T dongle and an SBC demonstrates that sophisticated localization can be achieved with inexpensive, compact hardware, making it practical for μAUVs.
- Passive, one-way communication is a significant advantage: The system's passive nature means it only listens to beacon signals, eliminating the need for active transmission from the μAUV and making it scalable for large μAUV fleets without increasing complexity per vehicle.
- Tank geometry significantly impacts accuracy: The experiments revealed that localization accuracy is highly dependent on the receiver's position relative to the beacons and the tank walls. Accuracy is best within the "convex hull" of the beacons and degrades near walls due to reflections.
- Trade-offs between cost/size and accuracy are inherent: While the embedded system achieves centimeter-level accuracy, it is an order of magnitude less precise than previous systems using full-fledged spectrum analyzers. This highlights the engineering trade-off between miniaturization/cost and absolute performance.
Key Results
- Static Position Estimation: The system achieved localization with a Particle Filter (M=1000 particles) at ten static positions. While the root-mean-square errors were higher (on the order of magnitude of centimeters) compared to prior work (millimeters), the authors attribute this to the hardware's processing capabilities and the tank's reflections. However, the measurements showed small variances, suggesting the underlying model is sound.
- Dynamic Position Estimation: Experiments with rectangular trajectories (both within and outside the convex hull of beacons) showed that both EKF and PF approaches provided similar results. The system demonstrated robustness, recovering position estimates even after temporary rejection of RSS measurements.
- Accuracy Dependence: Localization accuracy was found to be highest within the convex hull of the beacons and decreased with distance from beacons and near tank walls.
- System Cost and Size: The system is designed to be compact and low-cost, with the DVB-T dongle costing approximately USD 25 and the SBC (Raspberry Pi Zero) fulfilling the size and weight requirements for μAUVs.
Limitations and Future Work
- Accuracy limitations: The current accuracy is an order of magnitude lower than previous non-embedded systems, primarily due to the processing limitations of the DVB-T dongle and the reflections in the test tank.
- Tank geometry effects: The significant influence of tank walls and their reflections on localization accuracy is a major limitation.
- 3D Localization: The current system is primarily focused on 2D localization. Extending it to 3D is a significant challenge, especially due to antenna directivity.
- Future Work:
- Larger test environments: Deploying the system in larger water tanks (5-10m characteristic lengths) to achieve better EM field homogeneity and reduce wall reflection effects.
- Improved antenna design: Further investigation into antenna design to enhance performance.
- 3D localization: Exploring methods to achieve 3D localization, potentially by stacking beacons in multiple horizontal planes and fusing EM data with depth measurements from the μAUV.
- Integration into μAUVs: Embedding the system into μAUVs for closed-loop position control.
- Further analysis of EM field characteristics: Deeper investigation into the EM field properties of different underwater environments.
Summary updated