An adaptive model predictive control method for autonomous driving obstacle avoidance considering trajectory prediction uncertainty

Abstract

To address the issue that autonomous vehicles are susceptible to the uncertainty of future motion of surrounding vehicles in dynamic traffic environments, and that traditional Model Predictive Control (MPC) methods typically rely on deterministic trajectory prediction and struggle to dynamically adjust safety boundaries based on predicted risks, this paper proposes an adaptive MPC method that integrates trajectory prediction uncertainty. First, an ensemble-gated recurrent unit (Ensemble GRU) is used to model the historical trajectory of the target vehicle. The mean of the target vehicle's future trajectory prediction is obtained from the outputs of multiple GRU models, and the dispersion between the prediction results of each sub-model is used to characterize the trajectory prediction uncertainty. Second, the prediction uncertainty is incorporated into the obstacle avoidance safety constraint design, constructing an uncertainty-driven adaptive elliptical safety constraint that allows the safety boundary to automatically expand or contract based on the future motion risk of the target vehicle. Finally, this adaptive safety constraint is embedded into the MPC optimization framework, comprehensively optimizing vehicle trajectory tracking accuracy, control input smoothness, and dynamic obstacle avoidance safety. Simulation results show that the proposed method can accurately predict the future trajectory of the target vehicle and dynamically adjust the safety constraint range according to the prediction uncertainty, so that the autonomous vehicle can maintain a relatively stable driving state and a more reliable obstacle avoidance capability in complex dynamic scenarios, thereby improving the safety, robustness and environmental adaptability of motion control of autonomous vehicles.