An End-to-End autonomous driving model based on visual perception for temporary roads.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Deep learning for autonomous driving shows progress on structured roads, but limited research exists for temporary roads.
• End-to-End models offer high integration, translating input data directly to driving actions, enhancing safety and stability.
• Autonomous navigation in unstructured environments like temporary roads presents unique challenges.

Purpose of the Study:

• To propose a novel End-to-End autonomous driving model specifically for temporary roads, designed for mobile robots.
• To enhance the safety and stability of autonomous vehicles operating in challenging, unstructured environments.
• To enable unmanned ground vehicles to perform tracking tasks effectively on temporary roads.

Main Methods:

• A novel End-to-End model was developed, utilizing three road images as input.
• Image features were extracted using the Global Context Vision Transformer (GCViT) network.
• Local path planning was performed using Transformer and Gated Recurrent Unit (GRU) networks, with steering angle output via a control model.

Main Results:

• The End-to-End model demonstrated accurate identification of temporary roads.
• Single-frame trajectory planning time averaged approximately 100 ms.
• The average trajectory deviation was recorded at 0.689 m, meeting real-time processing needs.

Conclusions:

• The proposed End-to-End model is effective for autonomous driving on temporary roads.
• The model's real-time performance enables low-speed unmanned vehicles to execute tracking tasks in temporary road environments.
• This research contributes to advancing autonomous navigation capabilities in unstructured and dynamic road conditions.