Adaptive Delivery for High Definition Map Using A Multi-Arm Bandit Approach

Abstract

A high definition (HD) map is a key technology that enables autonomous driving, which has the characteristics of frequent updates and low latency requirements. Edge computing provides an efficient way to deliver the HD map to autonomous vehicles, which deploys the edge servers at the edge of the network and shortens the transmission distance. The edge-assisted HD map delivery is generally done by the wireless transmission between edge servers, like roadside units (RSU), and vehicles. However, the transmission channel status, like the transmission rate, is fragile and easily influenced by the speed of vehicles, the weather, and the number of connections of RSU. A proper HD map delivery is needed to meet a time deadline over different channel conditions. This work firstly utilizes the love-of-variety-based method to model the different versions of the HD maps with different data sizes. Then, an adaptive upper confidence bound based multi-arm bandit method is proposed to choose the appropriate version of the HD map under the different wireless communication statuses. The simulation results show the effectiveness of our proposed method, which achieves the best total accumulative rewards and the least regret compared with the baseline methods.