ZST-CBTM: Trajectory and Motion Prediction of Autonomous Vehicles Using Advanced Deep Learning Model

Trajectory prediction is crucial for autonomous vehicle systems to handle the route traffic and directly influences automated traffic safety. Recent techniques have attempted to learn to directly regress the exact future position or its distribution from massive amounts of trajectory data. However, there is uncertainty and uncontrollability in the process of trajectory prediction of surrounding obstacles increases the complexity of prediction. Consequently, this research proposes the Zero attention coupled Spatiotemporal features-based Convolutional Bi-LSTM (ZST-CBTM) method for meticulously predicting the trajectory and motion of autonomous vehicles. In the proposed approach, the Convolutional encoder is exploited to predict the features without change in the size of the input and enhances the output performance of the Bi-LSTM decoder. Specifically, the Bi-LSTM encoder effectively captures the temporal and spatial dependencies from the input data for attaining efficient prediction. In addition, the zero-attention module incorporated in the BiLSTM decoder enhances the performance of the classifier via introducing a zero vector that restricts the model from paying attention to unwanted input for processing. Moreover, the ZST-CBTM model extracts the global information features and local discriminative features, resulting in efficient performance in trajectory and motion prediction. The experimental results illustrate that the proposed ZST-CBTM model achieves less error values in terms of metrics MAE, MSE, RMSE, and R2 achieving 1.53, 5.33, 2.31, and 0.96 respectively outperforming other existing techniques.