A Comparison of the Performance of ResNet and DenseNet for Obstacle Avoidance and Path Following on a Mobile Robot Using Embedded GPUs

In this study, we evaluated the performance of deep learning-based navigation models on an novel platform that combines the Jetson Xavier NX with an RC four-wheel-drive car for autonomous driving. Focusing on data collection and model training using a single-camera setup, obstacle avoidance and path navigation were prioritised. The study compares the effectiveness of ResNet18, ResNet34, and DenseNet121 models by analysing training time, accuracy, obstacle avoidance efficiency, and path navigation capabilities. The results highlight the computational advantages of the Jetson Xavier NX over the Jetson Nano, demonstrating enhanced real-time processing and decision-making capabilities. This integration of the Jetson Xavier NX with an RC vehicle not only improves overall performance but also introduces a novel approach to autonomous vehicle development. The findings offer valuable insights into optimising deep learning models and hardware configurations for improved autonomous navigation.