Adaptive physics-informed trajectory reconstruction exploiting driver behavior and car dynamics

Makridis, Michail A.; Kouvelas, Anastasios · 2023 · Crossref

DOI: 10.1038/s41598-023-28202-1

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of reconstructing accurate vehicle trajectories from noisy, heterogeneous sensor data, a critical issue for traffic flow analysis, emissions estimation, and safety investigations. Existing methods typically rely on heuristic filtering strategies that prune or smooth data to achieve physically plausible values, but these approaches often lack adaptability to varying noise levels and sensor specifications. Furthermore, without ground-truth data, it is difficult to determine the optimal filtering strength, leading to potential distortion of microscopic driving dynamics. The authors propose an adaptive, physics-informed framework that integrates vehicle power dynamics and typical driver behavior constraints to automatically determine the optimal filtering magnitude, thereby ensuring reconstructed trajectories are both physically feasible and representative of real-world driving patterns. The proposed framework operates through an iterative process comprising three components: Vehicle Dynamics Constraint (VDC), Driver Dynamics Compliance (DDC), and Noise Reduction (NR). VDC and DDC utilize vehicle specifications (e.g., mass, torque, gearbox) and the Microscopic Fuel Consumption (MFC) model to define feasible acceleration boundaries based on speed, accounting for the fact that human drivers rarely utilize full vehicle acceleration capacity. The NR component applies standard filtering techniques—such as Moving Average, Lowess Polynomial Regression, or Butterworth filters—but automatically adjusts their strength (e.g., window size or cut-off frequency) by minimizing local acceleration variance under stable conditions. This threshold-free approach ensures the filtering is invariant to the quality of the input data. The framework was validated using synthetic data derived from high-accuracy differential GPS references with five levels of added Gaussian noise, as well as real-world data collected simultaneously by two smartphones with differing hardware specifications. In synthetic tests, the adaptive framework significantly outperformed fixed-parameter filtering methods, reducing speed error by up to 80% and preserving microscopic acceleration dynamics that static filters often distorted. In the real-world campaign, the framework reduced the mean absolute error between the two disparate device signals by 22.7% for speed and 69.7% for acceleration, effectively harmonizing the trajectories despite significant differences in sensor noise and sampling variance. The significance of this work lies in its ability to enable objective comparisons between drivers and vehicles equipped with different sensing technologies. By enforcing compliance with realistic nonlinear vehicle and driver dynamics, the method provides a robust, model-agnostic solution for trajectory reconstruction that does not require ground-truth validation for parameter tuning. This facilitates more reliable analyses in fields such as energy consumption modeling, driver behavior identification, and traffic safety, particularly in large-scale campaigns where sensor heterogeneity is common. The authors note that while the current model assumes known vehicle specifications, generic models can be substituted for large-scale applications, and future work may extend the framework to include lateral dynamics.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• crash reconstruction hf