Evaluating the Safety Impact of Network Disturbances for Remote Driving with Simulation-Based Human-in-the-Loop Testing

Abstract

One vital safety aspect of advanced vehicle features is ensuring that the interaction with human users will not cause accidents. For remote driving, the human operator is physically removed from the vehicle, instead controlling it from a remote control station over a wireless network. This work presents a methodology to inject network disturbances into this communication and analyse the effects on vehicle manoeuvrability. A driving simulator, CARLA, was connected to a driving station to allow human-in-the-loop testing. NETEM was used to inject faults to emulate network disturbances. Time-To-Collison (TTC) and Steering Reversal Rate (SRR) were used as the main metrics to assess manoeuvrability. Clear negative effects on the ability to safely control the vehicle were observed on both TTC and SRR for 5% packet loss, and collision analysis shows that 50ms communication delay and 5% packet loss resulted in crashes for our test setup. The presented methodology can be used as part of a safety evaluation or in the design loop of remote driving or remote assistance vehicle features.