TL;DR
This paper introduces a novel "Distance Increment" (DI) smoothing method to improve the accuracy of cooperative positioning for multi-platform systems. By leveraging the DI from dead reckoning to smooth the DI from cooperative positioning, the method effectively reduces random noise and detects/estimates Non-Line-of-Sight (NLOS) errors, leading to more stable and reliable formation control.
What Problem Does This Paper Address?
Cooperative missions involving multiple unmanned platforms (like drones or robots) rely heavily on accurate relative distance information for positioning and coordinated movement. However, real-world environments present challenges:
- Random Noise: Sensor measurements are inherently noisy, degrading positioning accuracy.
- Non-Line-of-Sight (NLOS) Errors: Obstacles can block direct communication paths, causing significant errors in distance measurements (e.g., from UWB sensors).
Existing methods for mitigating these errors often struggle with dynamic environments, require specific platform layouts, or are computationally intensive. This paper aims to address the need for a robust, flexible, and effective method to reduce distance errors, particularly NLOS, in cooperative positioning scenarios.
Methods
The core of this paper is the Distance Increment (DI) smoothing method. Here's how it works:
- Distance Increment (DI) Calculation:
- Dead Reckoning DI: Calculated using inertial navigation systems (INS) and odometers on individual platforms. This provides a continuous, albeit potentially drifting, estimate of distance change.
- Cooperative Positioning DI: Calculated from the relative positions of platforms after cooperative positioning updates. This is generally more accurate but can exhibit sudden jumps due to position corrections.
- DI Smoothing: The paper proposes using a Kalman filter to smooth the DI. The dead reckoning DI is used as the predicted value, and the cooperative positioning DI is used as the observed value. A key innovation is treating the difference between these two DIs (the "DI deviation") as an additional unknown parameter in the Kalman filter. This allows the filter to dynamically estimate and compensate for the jumps in the cooperative positioning DI, resulting in a stable DI with small errors.
- Distance Filtering: The smoothed DI is then used in a separate filtering model (also based on Kalman filtering) to smooth the measured distance itself. This step further reduces random noise in the distance measurements.
- NLOS Detection and Estimation:
- Detection: The smoothed DI is compared against a threshold (γmax). If the difference between the observed distance increment and the smoothed DI exceeds this threshold, an NLOS event is suspected.
- Estimation: Once an NLOS is detected, the method estimates the NLOS error by analyzing a sequence of observed distances and their corresponding estimated distances over a short time window (M). The mean of these differences is taken as the estimated NLOS error.
Datasets
The paper describes experimental setups but does not explicitly mention using pre-existing public datasets for training or evaluation. Instead, they conducted their own experiments with three mobile platforms.
Key Insights and Ideas
- DI as a Stabilizer: The DI, particularly when smoothed using dead reckoning, provides a stable signal that can effectively "anchor" the more accurate but jumpy cooperative positioning DI. This is crucial for robust filtering.
- DI Deviation as a Key Parameter: Treating the difference between dead reckoning DI and cooperative positioning DI as a dynamic parameter in the Kalman filter is a novel way to handle the inherent discrepancies and jumps in cooperative positioning updates.
- Unified Approach for Noise and NLOS: The DI smoothing method serves a dual purpose: it first smooths the DI itself, and then this smoothed DI is used to filter the raw distance measurements and detect/estimate NLOS errors. This integrated approach simplifies the overall processing pipeline.
- Adaptive NLOS Estimation: The method estimates NLOS errors based on a sequence of measurements, making it more robust to the transient nature of NLOS events.
Key Results
- DI Smoothing: The DI obtained through the proposed method exhibits small errors and no large jumps, combining the advantages of both dead reckoning and cooperative positioning DIs.
- NLOS Estimation Accuracy: In simulation, the DI-based method achieved a relative accuracy of NLOS estimation of 8.17%, significantly outperforming a reference method (RN-Based) which had errors up to 20.62%.
- Random Error Reduction: The maximum random error was reduced by 40.27%, and smoothness improved by 20.32% in simulation.
- Experimental Validation: In real-world experiments, the DI-based method successfully detected and modified NLOS errors. For segments with NLOS, the maximum error was reduced by up to 66% compared to the RN-Based method.
- Improved Positioning Stability: The DI-based method resulted in fewer fluctuations and more stable positioning results, enhancing the reliability of cooperative positioning.
Limitations and Future Work
- NLOS Simulation: The paper acknowledges that simulating accurate NLOS error distribution is challenging, and they relied on adding numbers obeying normal distribution based on literature.
- Threshold Setting: The effectiveness of NLOS detection depends on the γmax threshold, which needs careful tuning. The paper suggests adaptive windowing estimation if ranging accuracy is unknown.
- Future Work: The authors plan to focus on:
- Detecting NLOS from the signal level.
- Developing relevant fault detection strategies.
- Providing cooperative positioning services for multiple unmanned platforms in complex environments.
Summary updated