Multiple vehicle trajectory prediction method based on long-short memory network

Abstract

The invention discloses a multiple vehicle trajectory prediction method based on a long-short memory network, and the method comprises the steps of taking the historical trajectories of a main vehicleand an adjacent vehicle as the input, and fully considering the mutual impact between the positions of the vehicles and the driving behaviors; encoding and decoding through a network; further inputting the outputted future track of the adjacent vehicle into a hybrid density network; estimating the probability distribution of the vehicle positions; at every training, combining the error of a hybrid density network, a root-mean-square error of the trajectory result and the parameter regularization item of an encode-decoder network to form a loss function; and guiding the update of the network parameters, so that the prediction accuracy of the network can be improved. According to the method, the trained neural network can predict the probabilistic position information of the adjacent vehicles, and the position information forms a continuous track according to the time sequence, so that the main vehicle can be assisted to make decisions and plan.

Description

technical field [0001] The invention relates to the technical field of automatic control, in particular to a trajectory prediction method for multiple vehicles based on a long-short memory network. Background technique [0002] Autonomous driving and related research have made great progress in the past few decades. However, achieving a high level of autonomous driving in complex urban environments such as highways still faces great challenges. This is because in this environment, the driving behavior of unmanned vehicles is highly dynamic, and the driving strategies and positions of different road participants will be affected by each other. The estimation of the location of the vehicle will lead to overly aggressive or conservative driving behavior. For example, a vehicle in an adjacent lane suddenly changes lanes, which is likely to cause a rear-end collision without prediction and precaution. [0003] In this case, it is necessary to predict the future trajectories an...

AI Technical Summary

Problems solved by technology

Such a system will lead to the following problems: 1. It does not conform to the limited view angle of the sensor of the unmanned vehicle in the actual driving process. Generally speaking, the main vehicle driving on the road can only obtain the position information of the adjacent vehicles around it through the sensor. ; 2. A prediction result only includes the future trajectory of one vehicle, which is inefficient; 3. It ignores the interaction between vehicle driving behaviors within a certain range of vehicle clusters, making the prediction results unable to be applied to actual situations

Images