The effect of latency, speed and task on remote operation of vehicles

Abstract

IntroductionAlthough self-driving vehicle technology has been developing rapidly in recent years, there are still many challenges left before full autonomy can possibly be achieved. Remote operation could facilitate the development of autonomous vehicles in a safe and efficient manner by putting a human in the loop without the need for the human to be physically present in the vehicle. In the current experiment, three aspects of remote driving have been investigated in a driving simulator to evaluate the effect of i) latency, ii) type of task to perform, and iii) speed on a number of performance measures. MethodThirty-one participants drove in simulated rural (high-speed) and urban (low-speed) scenarios. Five hazards were created for each scenario and the participants drove each scenario three times with different latencies (baseline, +100 ms, and +200 ms). The latency condition was masked for the participants. The hazards were designed with the intention of creating challenging traffic situations. For example, in hazard one (H1) a car parked next to the road activates their turn indicators and then cuts into the participant’s lane close in front of the ego vehicle, forcing the participant to either brake or veer. Latency, type of hazard, and scenarios (high- and low-speed) were all within participants’ variables. Objective simulator data collected included variables such as reaction time, post-encroachment time, speed variation, distance to hazard, collisions, etc. Subjective data was gathered through questionnaires between each of the balanced latency conditions to assess trust, perceived control, realism of scenarios, and workload etc. After the completed drive, participants were asked to rate in which order they believed they had been subjected to the different latencies. The participants were divided into two groups, experienced drivers and experienced gamers. ResultThe results of the simulator study show that for some of the hazards, but not all, there were significant differences in the latency conditions and there were interaction effects between participant groups and environment/speed. For example, in H1 the effect on the reaction time was significantly larger than the added latency. Overall, the experienced gamers drove with larger safety margins although they had not been told that the latency was varied. Speed, latency, and group characteristics were interacting in significant ways and affected performance measures. The subjective ratings show that participants experienced less control of the vehicle during higher latency conditions, even though they were not told in which order they had been subjected to the latency conditions. The separate tasks to perform were affected differently by the independent measures. The number of collisions was not affected by latency. ConclusionThere seems to be a certain level of adaptivity among the participants, although they were not told that the latency varied between scenarios, and they could also not guess in which order they drove with the different conditions. In some situations, they drove with larger safety margins, especially experienced gamers in the high-speed scenario. Moreover, the subjective ratings show that participants felt less in control of the vehicle at higher latencies without being able to pinpoint what it was that affected their driving.