Uncertainty-Aware Path Planning With Multiple Constraints for Autonomous Vehicles Based on Nonlinear Reduced-Order Model Predictive Control

In this paper, we present a Risk-Aware Nonlinear Reduced-Order Model Predictive Control framework that utilizes real-time state estimation, reduced-order modeling, and optimization for reliable and efficient path planning with low energy consumption. Our method employs filtering techniques to estimate the system’s state by correcting for process and measurement noise, providing an accurate representation of the vehicle’s position and environment. A reduced-order model predicts future states based on the estimated input, while the optimizer adjusts control inputs to meet trajectory and safety constraints, such as avoiding obstacles. By incorporating adapted filtering techniques and optimized cost functions, our approach minimizes computational overhead while maintaining precise control over vehicle navigation. In the experiments, we introduce communication and process noises, allowing the robot to operate without knowing the precise location of obstacles and resulting in potential near-collisions. This involves redefining the constrained optimal control problem for a reduced-order model, addressing input constraints amidst uncertainties. Our method achieves over 94% success in collision-free navigation within the first 50 steps, while maintaining precise trajectory alignment and adapting to environmental changes. It also improves prediction speed up to 15% per cycle compared to NMPC and demonstrates 91% overall effectiveness in avoiding collisions and reaching the target.