Evaluation Framework for Trajectories Predicted in Autonomous Vehicle Traffic Prediction

摘要

According to one embodiment, when a prediction trajectory is received, one or more feature sets are extracted from at least some of the trajectory points of the prediction trajectory. The prediction trajectory is predicted using a prediction method or algorithm based on cognitive data for recognizing an object in a driving environment around an autonomous vehicle (ADV). The extracted feature is fed to a given DNN model to generate a similarity score. The similarity score represents the difference or similarity between the prediction trajectory and the previous actual trajectory used to train the DNN model. The similarity score can be used to evaluate the prediction method that predicted the prediction trajectory.