TL;DR
This paper introduces a novel "plane-constrained RTK" (PCRTK) method that leverages cooperative vehicle data to improve high-precision positioning in challenging urban environments. By fitting a road plane using data from nearby vehicles, PCRTK enhances the accuracy and reliability of traditional RTK, especially where GNSS signals are frequently blocked.
What Problem Does This Paper Address?
Achieving centimeter-level positioning accuracy is crucial for applications like autonomous driving, and Real-Time Kinematic (RTK) is a leading technology for this. However, RTK performance degrades significantly in urban canyons due to signal blockages from skyscrapers and overpasses. This leads to unreliable positioning, which is unacceptable for safety-critical systems. While fusing data from other sensors (like IMUs or LiDAR) can help, these methods often require expensive hardware. This paper addresses the need for a more robust and cost-effective solution for urban positioning by exploiting the inherent geometric constraints of vehicles traveling on roads.
Methods
The core innovation is the Plane-Constrained RTK (PCRTK) method. It works by:
- Road Plane Construction: Cooperative vehicles share their positioning data. This data is used to fit a "road plane" that represents the surface on which these vehicles are traveling. This is done by first fitting a plane to the vehicles' positions after removing antenna height, and then reintroducing the antenna height to define the vehicle plane.
- Plane Parameters as Measurements: The parameters of this fitted plane are then treated as new, albeit indirect, measurements within an adaptive Kalman filter.
- Adaptive Kalman Filter: The Kalman filter is adapted to fuse these plane-derived measurements with traditional GNSS observations. An "innovation-based adaptive Kalman filter" is used to dynamically update the noise covariance matrices (Q and R), making the estimator more robust to environmental interference like multipath.
- Integer Ambiguity Resolution: The standard LAMBDA method is employed for resolving carrier-phase ambiguities to achieve the final fixed, high-precision solution.
Datasets
The paper utilizes GNSS data collected from a four-vehicle fleet in both open-sky and dense urban environments in Beijing, China. The data was collected using various consumer-grade and professional GNSS receivers. For simulations, reference solutions were obtained using the NovAtel Inertial Explorer software, which processed raw GNSS observations along with INS data.
Key Insights and Ideas
- Exploiting Road Geometry: The fundamental insight is that vehicles on a road are constrained to a relatively smooth plane. This geometric constraint, often overlooked, can be a powerful source of information to augment GNSS positioning.
- Cooperative Data for Geometric Constraints: By leveraging data from multiple vehicles, the system can construct a more robust estimate of this road plane, even if individual GNSS signals are weak or intermittent.
- Plane Fitting as a Pre-processing Step: The accuracy of the fitted plane is critical. The paper introduces a plane detection algorithm to filter out inaccurate positioning solutions from cooperative vehicles, preventing them from degrading the plane estimate.
- Adaptive Filtering for Robustness: The use of an adaptive Kalman filter is crucial for handling the noisy and uncertain measurements that arise in urban environments, especially when dealing with multipath and signal dropouts.
Key Results
- Improved Accuracy in Urban Areas: Simulations and experiments showed that PCRTK significantly improves positioning accuracy compared to traditional RTK in urban environments. For example, the probability of positioning errors less than 1m increased from 88.48% to 96.76% for one vehicle in open-sky simulations.
- Reduced RMSE: In degraded urban environments, the RMSE for Vehicle 2 dropped from 6.74m (RTK) to 2.14m (PCRTK) horizontally, and from 14.93m to 2.36m vertically.
- Cooperative Network Benefits: When all vehicles in the network used PCRTK, the performance further improved. RMSE for Vehicle 4 decreased from 1.01m (PCRTK without cooperative network) to 0.63m (PCRTK with cooperative network).
- Robustness to Plane Uncertainty: While accuracy degrades with increasing plane uncertainty (simulated using Gaussian noise), the PCRTK method still outperformed traditional RTK up to a certain level of noise.
Limitations and Future Work
- Plane Assumption: The method relies on the assumption that vehicles are traveling on a relatively smooth plane. This might be limited in highly undulating terrain or rapidly changing road slopes.
- Cooperative Vehicle Accuracy: The accuracy of the fitted plane depends on the accuracy of the positioning data from cooperative vehicles. If these vehicles provide inaccurate data, it can negatively impact the plane estimate. The paper addresses this with a plane detection algorithm but acknowledges it as a potential limitation.
- Vehicle Separation: The optimal separation distance between cooperative vehicles for accurate plane fitting is not fully explored and is suggested for future work.
- Larger-Scale Networks: The performance with a larger number of cooperative vehicles needs further investigation.
- V2X Integration: The authors plan to explore employing roadside units (RSUs) for V2X scenarios to provide more reliable positioning information.
- Tighter Thresholds: For better integer ambiguity fixing, a tighter threshold for plane fitting is recommended, which will be evaluated with more experimental data.
Summary updated