Abstract: Understanding and quantifying the performance of sensing architectures on autonomous vehicles is a necessary step towards certification. However, once this evaluation can be performed, the combinatorial number of potential sensors on the vehicle limits the efficiency of a design tradespace exploration. Several figures of merit emerge when choosing a sensor suite; its performance for a specific autonomy task, its monetary cost, energy consumption, and contribution to the latency of the entire system. In this paper, we present formulations to evaluate a sensor combination across these dimensions for the localization and mapping task, as well as a method to enumerate architectures around the Pareto Front efficiently. We find that, on a benchmarked environment for this task, combinations with LiDARs are situated on the Pareto Front.
Presentation available HERE