Abstract
Mixed vehicle platoons, comprising connected and automated vehicles (CAVs) and human-driven vehicles (HDVs), hold significant potential for enhancing traffic performance. Most existing research assumes linear system dynamics and often ignores the impact of critical factors such as noise, disturbances, and attacks, which are inherent to real-world scenarios. To address these limitations, we propose a Robust Nonlinear Data-Driven Predictive Control (RNDDPC) framework that ensures safe and optimal control under uncertain and adverse conditions. By utilizing Koopman operator theory, we map the system's nonlinear dynamics into a higher-dimensional space, constructing a Koopman-based linear predictor that approximates the behavior of the original nonlinear system. To mitigate modeling errors associated with this predictor, we introduce a data-driven reachable set analysis technique that performs secondary learning using matrix zonotope sets, generating a reachable set predictor for over-approximation of the future states of the underlying system. Then, we formulate the RNDDPC optimization problem and solve it in a receding horizon manner for robust control inputs. Extensive simulations demonstrate that the proposed framework significantly outperforms baseline methods in tracking performance under noise, disturbances, and attacks.
Introduction: The Challenge of Mixed Vehicle Platoons
Mixed vehicle platoons—where a few CAVs are interspersed with many HDVs—offer a path toward smoother traffic flow and better fuel economy. However, controlling them is a significant challenge due to two main factors: the inherent nonlinearity of human driving behavior and the presence of environmental uncertainties.
Real-world systems are plagued by "adverse factors," including measurement noise, external disturbances (like the lead vehicle suddenly braking), and potential adversarial attacks (like false data injected into the control channel). Existing methods like Data-Enabled Predictive Control (DeePC) often struggle with noise or assume the system is linear. This paper aims to bridge the gap by creating a controller that is both data-driven and robust under pressure.
Methodology: The RNDDPC Framework
The RNDDPC framework consists of three main pillars designed to handle nonlinearity and uncertainty:
- Koopman Operator Theory: To handle nonlinearity, the authors use Koopman theory to "lift" the system's state into a much higher-dimensional space. In this higher dimension, the complex nonlinear dynamics of the platoon behave like a linear system. A Deep Neural Network (DNN) is used to learn this "lifting function" automatically from data.
- Data-Driven Reachability Analysis: Instead of predicting a single "best guess" for future states, the system calculates a Reachable Set. Using a geometric representation called matrix zonotopes, the controller identifies a "safety bubble" that accounts for modeling errors and noise. If the entire bubble satisfies safety constraints, the control input is considered robust.
- Receding Horizon Optimization: The controller solves an optimization problem at every time step, looking a few seconds into the future to find the best control inputs that keep the platoon within safe bounds while minimizing tracking errors.
Experimental Setup
The study relies on high-fidelity simulation data to validate the framework:
- Training Data: Generated using the Optimal Velocity Model (OVM) and Intelligent Driver Model (IDM) to simulate realistic human car-following behavior.
- Evaluation Environment: The authors used PreScan, employing a high-fidelity Audi A8 model with full two-dimensional dynamics, providing a more realistic testbed than standard academic "point-mass" models.
- Benchmarks: The system was tested against the WLTP (Worldwide harmonised Light vehicles Test Procedure), a standard driving cycle that includes urban, suburban, and highway phases.
Key Results and Insights
The RNDDPC was compared against several baselines, including Nonlinear MPC (NMPC), Linear MPC (LMPC), and DeePC.
- Superior Accuracy: In comprehensive simulations, RNDDPC achieved the lowest tracking errors. It reduced tracking velocity mean error by 3.6% and spacing error by 3.3% compared to the best-performing baseline.
- Extreme Robustness: Under cyber-attacks and noise, DeePC’s performance collapsed (with costs spiking 5x higher), while RNDDPC remained stable and efficient. By shifting from point-prediction to set-prediction, the controller becomes "adversary-proof" against bounded noise.
- Real-Time Feasibility: Despite the complex math, the average computation time was only 22ms per step, well below the 50ms threshold required for real-time vehicle control.
Limitations and Future Work
While the RNDDPC framework shows promise, the authors identify several areas for future research:
- Communication Delays: The current model assumes near-instant data transfer, which may not hold true in real-world 5G/V2X networks.
- Lateral Dynamics: The paper focuses on longitudinal control (speed and spacing); future work must integrate lateral control for steering and lane changes.
- Dynamic Topologies: The framework currently assumes a fixed platoon and does not yet account for vehicles joining or leaving the platoon in transit.
Summary updated