Autonomous Car Motion Prediction Based On Hybrid Resnet Model

Autonomous vehicle is a technology that will substantially transform the transportation industry. It can make the cars drive faster with fewer accidents. In an autonomous driving system, trajectory prediction provides insight into the future movements of other agents around the vehicle, helping it to make better trajectory planning for itself. In this paper, we proposed a deep-learning-based hybrid model method to generate three possible trajectories for any given agent in an urban driving setting. Our model is made up of two 3D convolutional neural networks thus is faster and easier to train than Long Short Trem Memory-based methods. The loss function is also customized towards the highly dynamic and ambiguous situation in urban driving. Our method outperforms other deep-learning-based models, reaching a loss of 22.45.