Autonomous vehicles rely on continuous perception of their surroundings to make safe and effective driving decisions. To achieve this, they use computationally complex perception systems that combine data from various sensors such as cameras, LiDAR, and radars. Regarding the detection of these targets, Luca Lusvarghi, a researcher at the Institute of Engineering Research (I3E) of the UMH, has created the first AI-based model that estimates, in real-time, the relevance of various objects for an autonomous vehicle.
Until now, existing work has addressed the identification and prioritization of objects in autonomous driving environments primarily through conventional AI techniques or costly manual annotations. The model proposed by Lusvarghi, under the supervision of professor Javier Gozálvez, uses causal learning, a cutting-edge approach that directly measures the impact of each object on the vehicle's driving decisions.
«Understanding object relevance allows an autonomous vehicle to focus on those that truly matter and greatly optimize the use of computational resources, especially in high-density urban scenarios,» explains Lusvarghi. This reduces computational load, improves decision-making efficiency, and lowers energy consumption, making driving more efficient and human-like.
This capability also enhances communication between vehicles. The AI models enable connected and autonomous vehicles to select and transmit only relevant objects, reducing the amount of exchanged data by up to 2.8 times compared to existing methods. This is crucial for large-scale deployment and could allow communication networks to support up to four times more vehicles.
«Identifying relevant objects can impact the entire connected and autonomous driving ecosystem,» concludes Lusvarghi. These relevance estimation models could pave the way for a new generation of more efficient and scalable vehicles.
The work was developed within the framework of the SemanticV2X project, funded by the European Commission through the Marie Skłodowska-Curie Actions (MSCA-PF) postdoctoral fellowship program in 2023.
