.png){kind=icon}
.png){kind=icon}
Return
Intelligent Decision Algorithm for Autonomous Driving Based on Convolution Neural Network and Deep Deterministic Policy Gradient
.png){kind=icon}
PRE
AI
DOI：10.1142/S0218126625503803.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This study aims to develop an efficient and accurate intelligent decision-making algorithm for autonomous driving by integrating convolutional neural networks (CNNs) with a deep deterministic policy gradient (DDPG) model. The algorithm first employs a CNN to extract and identify features from image data of driving scenes. These features are then input into the DDPG model, which performs decision-making and control through an actor-critic network architecture. To address the significant training fluctuations commonly associated with DDPG, a synthetic experience replay mechanism is introduced. Additionally, a more rational reward function is designed to overcome challenges in defining appropriate reward values. The algorithm's effectiveness and reliability in autonomous driving have been validated through both simulation and real-world driving experiments. The results demonstrate that the proposed CNN-DDPG-based algorithm achieves high obstacle recognition accuracy. Although recognition performance slightly declines as driving speed increases, the algorithm maintains stable overall performance and surpasses other baseline methods. These findings highlight the algorithm's superior ability to process complex environmental data and devise optimal driving strategies.
Keywords:
Convolutional neural network
deep deterministic policy gradient
autonomous driving
deep learning
reinforcement learning
Journal
IF:
1
Papers:
376
Citations:
2.3K