Automatic driving dynamic obstacle avoidance simulation test method

Abstract

The invention discloses an automatic driving dynamic obstacle avoidance simulation test method which comprises the following steps: step 1, selecting a simulation platform CARLA 0.8.4 version, step 2, simulating an automatic driving model based on visual information, and establishing a DSCIL model, step 3, extracting forward image features by the DSCIL model by adopting a 34-layer ResNet network, carrying out environment perception by adopting continuous four frames of forward images, and obtaining a 640-dimensional environment feature vector through a residual network and an LSTM network; step 4, solving a vehicle transverse control problem by using a branch decision network; step 5, controlling the vehicle speed through proportional-integral control, and solving the longitudinal control problem of the vehicle; and step 6, preprocessing the data, setting corresponding experimental parameters and test results, and analyzing the data. According to the method, the speed and the motion trail of a dynamic obstacle can be sensed through the characteristics of continuous multi-frame images in a dynamic environment, so that the obstacle can be well avoided, and an automobile can achieve good safe driving ability.

Description

technical field [0001] The invention relates to the technical field of automatic driving obstacle avoidance, in particular to a simulation test method for automatic driving dynamic obstacle avoidance. Background technique [0002] Self-driving cars, also known as unmanned cars or wheeled mobile robots, are smart cars that realize unmanned driving through computer systems. It has a history of several decades in the 20th century, and at the beginning of the 21st century, it shows a trend close to practicality. [0003] In the process of human driving, it mainly relies on visual information to determine the traffic conditions ahead, and relies on the speedometer to determine the state of motion of the car. Relying on the neural network model to imitate human driving behavior requires collecting human driving records and sensor information for supervised learning. In 2005, Lecun et al. constructed an end-to-end model DAVE with a 6-layer convolutional neural network, using super...

AI Technical Summary

Problems solved by technology

[0004] In fact, during the driving process, human beings can judge the movement trend and speed of dynamic obstacles based on the visual information of the past period of time to determine the driving strategy, which is indispensable for driving in real scenes. However, for the above model , it is difficult to obtain the motion state of the dynamic obstacle, and the obstacle cannot be avoided very well. Therefore, an improved technology is urgently needed to solve this problem existing in the prior art

Images