Edge acceleration of LiDAR frame transmission with in-network machine learning

In real-time vehicle perception scenarios, ensuring timely and stable transmission of LiDAR data between vehicles and the network edge is crucial for accurate object detection. However, the inherent variability of wireless links, coupled with the added impact of vehicle mobility, leads to inevitable packet loss and latency jitter, compromising both the timeliness and accuracy of vehicle perception. To address this challenge, we introduce a packet duplication mechanism on dual wireless links, improving LiDAR frame transmission performance. The solution is driven by an integrated In-Network Machine Learning module at a programmable edge device that dynamically detects performance degradation and controls packet duplication. Through practical implementation and extensive evaluation, it is demonstrated that the proposed packet duplication function can effectively address uncertainties in LiDAR frame transmission, while achieving 50% reduction in transmission times.