Automatic Parking Trajectory Planning Based on Recurrent Neural Network

Relying on a company's automatic parking project, This paper proposes a method which use TensorFlow training Recurrent Neural Network (RNN) to predict parking trajectory curves and determine the optimal parking trajectory. First, we established a parallel parking model with obstacles and gave the starting point of the parking and the range of obstacles. Then use the arctangent function to discretize the model parking trajectory and introduce the discretized data as initial sample data into TensorFlow for Recurrent Neural Network (RNN) training to generate discretized parking trajectory curve data. Finally, the data will be summed into the actual parking trajectory curve and compared with the experienced driver parking trajectory curve. It is found that the parking trajectory curve which obtained by the method proposed in this paper is highly integrated with the rich driver parking trajectory curve, and even exceeded in the obstacle avoidance function.

I. Introduction

With the development of industry, cars have gradually become an indispensable part of people's work and life. As a result, the number of cars has grown faster and faster. With the improvement of people's living conditions, the population has also increased, and urban construction facilities have also increased, This has led to a growing crowd of urban spaces. Then there are a series of problems such as traffic congestion and reduced parking space for cars.