Reducing Parameters by Neuroevolution in CNN for Steering Angle Estimation

Convolutional Neural Networks (CNNs) are becoming increasingly popular in autonomous driving. Researchers are focused on optimizing these models to work on smaller control devices, which can allow for more performance of vehicle steering. A lot of attention is being paid to reducing the number of parameters in a CNN so that it can predict steering angles more efficiently for autonomous vehicles. To tackle this challenge, we have adopted a neuroevolution approach, which is a fusion of neural networks and evolutionary algorithms. This approach can help us find models that are less complex but still have adequate predictive capacity and a lower number of parameters compared to existing networks. The results show that the neuroevolution approach was successful in producing an architecture with only 35,972 evaluated parameters. This is a significant reduction in the number of parameters compared to established models like PilotNet with 233,701 parameters, RestNet50 with 25,636,699 parameters, PSO-CNN with 1,335,758 parameters, VGG16 with 138,357,456 parameters, and VGG16-LSTM with 156,223,443 parameters. Despite having fewer parameters, our model’s performance is similar to some of the other CNNs.