Abstract: To address the problem that target vehicle trajectory prediction in complex traffic environments exhibits multi-modality and uncertainty, making it difficult for traditional model predictive control (MPC) to effectively utilize prediction information, a probabilistic risk prediction– based control method is proposed. The multi-modal trajectory prediction results are modeled in the probabilistic space to construct state probability distributions and extract their statistical characteristics. A confidence ellipsoidal risk envelope is generated and incorporated into the MPC framework as risk constraints. Experimental results demonstrate that the proposed method can effectively avoid collision risks while improving trajectory smoothness in complex scenarios, providing a feasible and generalizable solution for prediction-driven safety decision-making control of intelligent vehicles.

Key words: probabilistic space modeling; risk envelope; model predictive control; trajectory prediction; multi-modal distribution

摘要： 针对复杂交通环境中目标车辆轨迹预测呈现多模态与不确定性，传统模型预测控制 （MPC）难以有效利用预测信息的问题，提出一种基于概率风险预测的控制方法。通过在概 率空间对多模态轨迹预测结果进行建模，构建状态概率分布并提取统计特性，利用风险包络， 并将其引入 MPC 框架形成风险约束。实验结果表明，该方法在复杂场景下能够在保证轨迹平 滑基础上有效规避碰撞风险，为预测驱动的智能车辆安全决策控制提供了一种可行且具有推 广价值的解决方案。

关键词: 概率空间建模；风险包络；模型预测控制；轨迹预测；多模态分布