The edge as a new element of road traffic infrastructure
Modern autonomous vehicles rely on sensor data, which has limited visibility. Obstructions, curves and dense traffic make it difficult to detect objects at a distance. Edge-based solutions can solve this problem by combining data from multiple sources - autonomous vehicles and roadside units (RSUs). In this context, the concept of Connected Autonomous Vehicles (CAV) emerges, which cooperate to create a common world model. However, the key challenge is the information transmission time - the data must be up-to-date in order to be meaningful for traffic planning systems.
The Conductor solution introduces a new approach: instead of local fusion at the vehicle level, it creates an integrated world model from the perspective of a fixed point, e.g., a roadside unit (RSU). This allows for better scene recognition, especially in obscured areas where vehicles cannot see beyond their sensors.
Dynamic data selection for real time
The main challenge in edge systems is the timing constraint - data must be processed and delivered within a specific time interval, known as Age of Information (AoI). If the information is too late, it loses its value. In the case of Conductor, the system does not accept data from all vehicles, but dynamically selects those that will bring the greatest benefit. The key is an occlusion-aware selector: data from vehicles that detect objects in areas not covered by the RSU's view are preferred.
This means that the system not only collects data, but also intelligently decides which of them are most valuable. In combination with a time controller, which adjusts the number of data for fusion and the amount of trajectory prediction in each iteration, the solution can stay within the AoI time limits even under high load.
Enlarged imageClose zoomPrevious imageSimulation tests: effectiveness in real conditions
Research conducted on the CAV simulation infrastructure has shown that the Conductor solution meets the safety requirement related to Age of Information (AoI) even with the simultaneous participation of up to 31 autonomous vehicles. This is important because in real-world traffic conditions, the number of vehicles can be significant, and the system must operate reliably.
The data fusion values were close to those achieved by an ideal model (Oracle), which means that the system can effectively utilize available data. Moreover, its results were significantly better than those obtained with a random selection of vehicles for fusion - which shows that intelligent data selection has a real impact on the quality of the system.
Significance and limitations of the solution
The Conductor solution is not deployed in real vehicles or road infrastructure - it is a simulation-based demonstration. This means that its effectiveness in real-world conditions, taking into account network interference, hardware diversity, and unpredictable driver behavior, remains unconfirmed. Additionally, the system requires time synchronization and stable communication between vehicles and stationary units, which may be difficult in practice.
Despite this, the study shows an important direction of development: the future of autonomous systems lies not only in developing sensors on vehicles but also in intelligently utilizing infrastructure and cooperation between vehicles. Edge data fusion can become a key element of road safety and efficiency.
Significance and limitations of the solution
The Conductor solution, although based on simulation, opens new perspectives for the future of traffic. Its key innovation lies in using edge data fusion as an infrastructure element that not only increases safety but also improves logistical efficiency and reduces the risk of collisions in dense urban areas. Thanks to intelligent data selection based on occlusion awareness, the system can effectively compensate for the limitations of autonomous vehicle sensors, which is particularly important in urban environments where turns, buildings, and other obstacles hinder visibility. It is worth emphasizing that this approach may be crucial for the development of CAV (Connected Autonomous Vehicles) systems on an urban scale, where the number of vehicles and the complexity of the scene are high.
However, there are real limitations that must be overcome before commercialization. This requires not only a stable and low-latency communication network but also time synchronization between vehicles and stationary units (RSU), which in real-world conditions may be difficult due to changing network conditions. Additionally, hardware diversity, different communication standards, and potential interference can affect the quality of data transmitted to the edge. Therefore, although the simulation results are promising, their transfer to real-world conditions requires further research in high-fidelity test environments, as well as cooperation between vehicle manufacturers, infrastructure operators, and regulatory bodies. In the future, such solutions may become a standard for smart cities, but only after overcoming technical and organizational barriers.
