In this project, we intend to propose the adaptive prediction-horizon MPC, where the prediction interval is adjusted in real-time based on observed vehicle behaviors while the prediction steps are kept the same. The adaptive MPC strategy optimizes the control performance by fine-tuning the prediction interval online to suit the vehicle's dynamic behavior without compromising the overall computation cost.
The standard MPC's fixed prediction horizon might not always be optimal for all driving scenarios and conditions. In certain situations, like high-speed driving or sudden changes in road conditions, a longer prediction horizon might be beneficial to anticipate and respond to potential challenges. In contrast, during low-speed urban driving or stable conditions, a shorter prediction horizon might suffice, enabling quicker and more agile control responses.
In this project, an adaptive-prediction-horizon MPC is proposed. The prediction interval is adjusted in real-time based on observed vehicle behaviors, while the prediction steps are kept the same, which can extend the prediction horizon while maintaining the computational cost at the same level.
The effectiveness of the developed adaptive MPC was tested through practical experiments on the electric Equinox vehicle using a double lane change (DLC) input. The proposed adaptive MPC continuously monitored the yaw rate and made necessary adjustments to the prediction interval when the yaw rate approached or exceeded its limit. When the yaw rate neared the limit, the controller increased the prediction interval to provide the controller with more time to react and mitigate yaw rate overshoot. Conversely, when the yaw rate remained well within the limit, the prediction interval was reduced, focusing on shorter-term control actions to avoid undershoot in the yaw rate tracking. Remarkably, this adaptative MPC strategy is achieved without compromising the computational efficiency, allowing for real-time vehicle implementation.
