Accurately perceiving the surrounding environment is a fundamental requirement for autonomous navigation tasks, with traffic signal perception at intersections being of particular importance. In traffic-intensive scenarios, incorrect perception of environmental information can result in erroneous control behavior of autonomous vehicles, potentially leading to severe consequences. In this paper, we present an autonomous navigation model that combines scene attention and traffic light detection to ensure safe control in urban intersection scenarios. We integrate traffic light detection into our autonomous navigation model using an oracle approach. Additionally, our model incorporates an attention mechanism that dynamically focuses on crucial features within the scene based on vehicle driving parameters. In our experiments, we evaluated our proposed model in cross scenarios under two weather conditions. The results demonstrate that our model performs similarly to the baseline models in terms of driving performance at intersections in the CARLA simulator. Moreover, it exhibits fewer traffic light violations and faster response times. In the ablation experiment, we performed numerous tests on the traffic light detection module using simulated and real datasets to evaluate its detection capabilities. Concurrently, we visualized the attention map to assess the effectiveness of the attention mechanism.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    Oracle-Based Conditional Imitation Learning for Autonomous Navigation


    Contributors:


    Publication date :

    2024-06-24


    Size :

    1429054 byte




    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    English