Safety Aware Predictive Control Neural Network for Connected Automated Vehicle Operations

Existing studies investigate trajectory prediction and trajectory planning as two separate problems assuming an ideal scenario without prediction errors. However, prediction errors commonly accumulate as the prediction period increases. Thus, the safety performance of the CAV trajectory planned based on the inaccurately predicted trajectory of the surrounding vehicle cannot be guaranteed. Inspired by the common optimization nature of trajectory prediction and trajectory planning, this study proposes a safety-aware predictive control neural network for CAV operations. Instead of first predicting the preceding HV trajectory and then optimizing the CAV trajectory based on the prediction, the SPCNN integrates the preceding HV trajectory prediction and the subject CAV trajectory planning as one model. It optimally adjusts the predicted trajectory to derive the CAV trajectory using a car-following model-based recurrent neural layer to ensure safety concerning the preceding HV’s actual trajectory and at the same time to maintain mobility.