Ensuring the safe decisions of autonomous vehicles (AVs) is of utmost importance, given their increasing popularity in public roads today. Sequential metamorphic testing has been successfully applied to reveal failures in various AV systems by verifying sequences or groups of metamorphic relations (MRs) among multiple of inputs and outputs. Having said that, new MRs are still required to address the complexity of driving scenarios. In this paper, we present diverse sequences of MRs for testing perception systems in AVs. Systematic applications of these MRs allow us to detect more failures from various deeplearning-based AV systems in response to diverse road objects, environment factors, and their combinations. The new MRs also help us gain more insights into these systems. The success of our study demonstrates that the journey to identify new MRs for testing autonomous driving is challenging yet significant and rewarding. Index Terms: Metamorphic Testing, Metamorphic Relations, Autonomous Vehicles

* This project is supported in part by the Discovery Grant DP210102447 of the Australian Research Council. 1. (Corresponding author.) Swinburne University of Technology, Melbourne, Australia. Email: 2. Swinburne University of Technology, Melbourne, Australia. 3. School of Computing and Data Science, The University of Hong Kong, Pokfulam, Hong Kong. Email:  EVERY VISITOR COUNTS: