PAINet: An Integrated Passive and Active Intent Network for Digital Twins in Automatic Driving

In recent years, the rapid development of artificial intelligence (AI) has accelerated the convergence of digital twin technology with the landscape of 6G automatic driving which is poised to exert profound and far-reaching effects on various aspects of human life. For example, it can offer solutions to prevent traffic accidents caused by blind spots, sudden malfunctions, and other factors. However, current digital twins used in automatic driving fail to meet the requirements for effective operation. This is mainly because the single view of vehicles is limited due to visual blind spots for object detection, while the information sensed by the network requires additional communication time, resulting in a lack of real-time updates and accurate labeling. To advance the level of digital twinning, this article proposes an integrated passive and active intent network (PAINet). This pioneering approach for communication amalgamates passive information sensed by vehicles and their behavioral data within vehicular networks. Through the extraction and detection of pertinent data at the intent layer, informed driving decisions are facilitated, particularly in response to unforeseen emergencies. Therefore, PAINet engenders a unified intelligent entity that seamlessly integrates communication, perception, computation, and control, ushering in a revolutionary safety paradigm for 6G automatic driving. We thoroughly constructed a vehicle intent dataset derived from conventional object detection datasets. In the context of our use case scenario, experimental findings strongly confirm the effectiveness of this methodology, revealing an impressive 96 percent precision in vehicle intent parse, along with a significant reduction of over 50 percent in processing delay. This innovative approach shows great potential for making significant strides in tackling the challenges of perception and decision-making in the realm of 6G automatic driving. We have uploaded the code to github.com/YU-spec-arch/PAINET.