TL;DR
This paper introduces a novel method for assessing the credibility of digital twins used in vehicle-in-the-loop (VIL) testing. By leveraging information entropy, specifically Approximate Entropy (ApEn) and Cross-ApEn, the authors quantify the "confusion" and "relevance" within the data streams of real and virtual vehicles, providing a more robust evaluation of VIL test systems. This is crucial for the reliable development and validation of increasingly complex intelligent vehicles.
What Problem Does This Paper Address?
The rapid advancement of intelligent and autonomous vehicles necessitates rigorous testing and validation. Vehicle-in-the-Loop (VIL) testing, which integrates real vehicles with virtual environments (digital twins), offers a promising solution for complex testing scenarios. However, a significant challenge lies in assessing the credibility of these dynamically evolving digital twin models within VIL tests. Traditional methods often fall short, especially when dealing with the continuous data streams and intricate interactions inherent in these systems. This paper aims to fill this gap by proposing a quantitative method to evaluate how reliably a digital twin VIL test system reflects the real-world performance of an intelligent vehicle.
Methods
The core of this paper is the development of a credibility assessment methodology based on information entropy. The authors propose using:
- Approximate Entropy (ApEn): This metric quantifies the complexity and regularity of a single time series. In this context, it's used to measure the "information confusion" within the data streams of either the real vehicle or the digital twin.
- Cross-ApEn (cross-approximate entropy): This extends ApEn to measure the similarity and correlation between two time series. Here, it's employed to assess the "structural relevance" or how well the data from the digital twin aligns with the data from the real vehicle.
The credibility of the VIL test system is then calculated as the difference between the ApEn of the system's own data and the cross-ApEn between the real and digital vehicle data. This difference, normalized, provides an "Information Entropy" score. The authors also introduce a "Credibility Evaluation Index" (ICR) derived from these entropy values to represent the overall credibility.
Datasets
The experiments were conducted using data collected from a VIL test system. The paper describes two specific scenarios:
- Scenario A: A stationary obstacle (vehicle) and a test vehicle approaching at a constant speed.
- Scenario B: A stationary dummy obstacle and a test vehicle approaching at a constant speed.
For both scenarios, the VIL test system and practical road tests generated data including the velocity of the test vehicle and the relative distance between the test vehicle and the target obstacle. This data was collected at a high frequency (10 ms) using GNSS/INS integrated navigation systems. The paper does not mention the use of publicly available datasets; rather, it relies on data generated from their experimental VIL setup and corresponding real-world tests.
Key Insights and Ideas
- Information Entropy as a Credibility Metric: The paper successfully demonstrates that information entropy, specifically ApEn and cross-ApEn, can be effectively used to quantify the credibility of digital twin VIL tests. This moves beyond simple data consistency checks to a more nuanced understanding of information flow and complexity.
- Inconsistent Weighting is a Major Factor: A critical finding is that the "inconsistent weight of the real and digital vehicle" is a significant factor impacting credibility. This likely refers to discrepancies in how the real and virtual systems are modeled or how their data is weighted, particularly highlighting the importance of factors like the number of people in the vehicle.
- ApEn and Cross-ApEn Capture Different Aspects: The distinct roles of ApEn (internal data confusion) and cross-ApEn (inter-system relevance) are crucial. Their combination provides a more holistic view of the VIL system's fidelity.
- Robustness to Series Length: The study emphasizes that the credibility assessment is relatively insensitive to the length of the data series (N), with variations in credibility fluctuating by no more than 2% even with significant changes in N. This suggests the method is robust for practical applications.
Key Results
- Scenario A: The credibility of the digital twin VIL system was found to be 75.68%, with a mean square deviation of 4.00%. The practical road tests showed less information entropy fluctuation.
- Scenario B: After adjusting for factors like vehicle weight, the credibility of the digital twin VIL system significantly increased to 96.12%, with a mean square deviation of 2.87%. This highlights the impact of accurate modeling of real-world conditions.
- Impact of Series Length: The analysis showed that increasing the length of the data series (N) had a minimal effect (no more than 2%) on the credibility assessment results, indicating the method's stability.
Limitations and Future Work
- Inconsistent Weighting: The paper identifies "inconsistent weight of the real and digital vehicle" as a key factor affecting credibility, particularly the number of people in the vehicle. This suggests that more sophisticated modeling of such real-world variations is needed.
- Future Work: The authors propose further analysis of other parameters influencing credibility in VIL tests based on digital twins. They also suggest exploring object classification approaches using information entropy within machine learning techniques for autonomous vehicles.
Summary updated