Motion Prediction for Autonomous Vehicles Based on EfficientNet-B1

Abstract

Perception, decision-making, and planning and control are the three cores of autonomous vehicles, enabling the vehicle to perceive and understand the driving environment, make active decisions, perform real-time path planning on the global or local map, accurately control vehicle motion, and track the desired trajectory. In this way, the desired driving requirements are achieved. In our paper, we use EfficientNet to predict autonomous vehicles' motion. We calculate the negative log-likelihood of the ground truth data given the multi-modal predictions. to evaluate our model's performance, we compare our model with ResNet50 and ResNet34 using same loss and same dataset. From the compared experiments, EfficientNet method owns the lowest loss 11.37. In contrast, the other methods like ResNext50, ResNet34's accuracy are 23.58, 23.62 respectively.