CoDe: Customizing Urban HD Map Deployment Strategy with Spatio-Temporal GPS Trace

Xiaofeng Cao,Deke Guo,Feng Lyu,Peng Yang,Weiming Zhang
DOI: https://doi.org/10.1145/3689823
2024-01-01
ACM Transactions on Sensor Networks
Abstract:In this paper, we investigate the edge cache deployment for high definition (HD) urban map provisioning, which is an essential building block for future autonomous driving. Given a deployment budget, we first formulate a satisfied downloading request maximization (SDRM) problem to obtain the deployment locations and customized cache sizes. The SDRM problem is unsolvable directly as the urban transportation traffic is highly dynamic and future traffic conditions are unknown in advance. Based on data analytics of the vehicular GPS trace, we propose an architecture named CoDe, built on which we transform and address the SDRM problem. The reference implementation of CoDe is respectively developed based on the insights from two urban-scale carpool GPS traces. The novelty and contributions of CoDe lie in its three-layer design. Particularly, at data feeding layer, we make use of two urban 60-day GPS traces involving respectively 37,801 and 17,517 vehicles, to extract the analytics samples. At mobility characterization layer, we conduct extensive data analytics on traffic mobility in terms of their distribution, correlation, and variation, to mine the crucial traffic mobility patterns for strategy customization. At cache deployment layer, we propose the K-order subgraph for each block to record the moving statistics within its K-order neighborhood, and transform the SDRM problem accordingly. Then, the Route weIght baSed grEedy (RISE) algorithm is devised for the problem, which can deliver the deployment decisions. Extensive data-driven experiments are carried out to demonstrate the superior performance of CoDe in terms of request hit ratio and caching resource utility.
What problem does this paper attempt to address?