Introduction
Accurately sensing vehicle dynamics in real-time is a cornerstone of automotive engineering, particularly for improving ride quality and safety. However, many critical states, such as vertical wheel velocity or body pitch, are difficult to measure directly with physical sensors. While mathematical estimation is a standard workaround, the computational complexity of high-fidelity "full-car" models often exceeds the capabilities of standard onboard Electronic Control Units (ECUs).
This paper, published in 2017, proposes a vehicle-to-cloud-to-vehicle (V2C2V) framework to offload these heavy calculations to a remote server, enabling advanced state estimation without taxing onboard hardware.
The V2C2V Framework
The core of the proposed solution is the V2C2V architecture. In this setup, the vehicle acts as a data collector and actuator, while the "cloud" serves as a high-performance remote agent.
As shown in Figure 1, the vehicle sends basic sensor data and GPS coordinates through a communication network to the remote agent. This agent performs complex Moving Horizon Estimation (MHE) and returns the estimated states to the vehicle's controller. A unique feature of this system is the integration of "Road Info" stored in the cloud—such as known potholes or terrain maps—which provides prior knowledge to the estimator, significantly enhancing accuracy beyond what a standalone vehicle could achieve.
Full-Car Modeling and Moving Horizon Estimation
To achieve realistic results, the researchers moved beyond the simplified "quarter-car" models common in literature, instead utilizing a 7-degree-of-freedom (DOF) full-car model.
This model, illustrated in Figure 2, accounts for vertical motion, pitch, roll, and the independent movement of all four wheels, resulting in 14 distinct state variables. To track these variables, the authors employed Moving Horizon Estimation (MHE). Unlike traditional Kalman Filters that process data point-by-point, MHE optimizes over a "window" of recent data. This allows the system to handle physical constraints (like suspension travel limits) and non-Gaussian road noise more effectively than simpler filters.
Simulation and Results
The framework was validated through numerical simulations using standardized vehicle parameters (e.g., a 1200kg sprung mass) and modeled road disturbances. The results demonstrated high fidelity across all 14 variables, with state estimates almost perfectly tracking the actual simulated states.
The system was particularly effective at tracking vehicle "attitude"—pitch and roll angles—which are vital for rollover prevention and passenger comfort. Even when faced with unknown disturbances or initial state errors, the estimator converged to the actual values rapidly, typically in less than a second.
Challenges and Future Directions
While the simulation results are promising, the authors highlight that real-world implementation faces the hurdle of network latency. The current model assumes a near-instantaneous connection; however, cellular latency and packet loss could delay the arrival of state estimates. Future research aims to integrate "state predictors" to maintain safe operation during temporary connection drops and to close the loop by applying these estimates to active damper control.
Summary updated