Cost-Effective Redundancy Approach for Fail-Operational Autonomous Driving System

Abstract

Driverless autonomous driving systems require cost-effective architecture satisfying design diversity and real-time performance to fulfill the fail-operational requirements that sustain system safety if a failure occurs during automated driving. However, conventional approaches cannot be applied to the systems due to design diversity. A key challenge in establishing a cost-effective multi-mode architecture is how to enhance the real-time capability of the mode switch. In this work, we propose three replication methods for fail-operational autonomous driving systems with design diversity: Input Backup Replication (IBR), Extended Primary Backup Replication (E-PBR), and Extended Leader Follower Replication (E-LFR). These methods enable accelerated recovery processing by utilizing input data and internal state backup in addition to partial hot standby. We implemented an autonomous driving prototype and found that (i) the proposed replication methods can satisfy the performance requirements for fail-operational systems, (ii) they can reduce 53.8 % of the CPU load compared with the hot standby approach in the normal mode, and (iii) the memory consumption ratio caused by the proposed methods is 0.01%. These results demonstrate that our proposed replication methods are feasible for fail-operational autonomous driving systems with design diversity.