Trajectory Prediction for V2X Collision Warning Using Pruned Transformer Model

This study explores the use of a pruned Transformer model for vehicle trajectory prediction in V2X collision warning systems, focusing on balancing accuracy and computational efficiency. By employing weight pruning, the model significantly reduces parameters and inference time while maintaining competitive performance with traditional LSTM models. Real-world experiments across diverse traffic scenarios demonstrate the pruned Transformer’s strengths in trajectory prediction error and false warning rates. However, increased missed warnings in dynamic scenarios and longer prediction times highlight areas for further optimization. These findings establish the pruned Transformer as a promising candidate for real-time V2X applications under resource constraints.