.png){kind=icon}
.png){kind=icon}
Return
Is this the real driver, or is it just a robot? A study of human-like autonomous driving controllers in a driving simulator
.png){kind=icon}
PRE
AI
F
S
T
T
DOI：10.1016/j.trf.2026.103619.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Autonomous vehicles (AVs) that exhibit human-like behavior have the potential to improve user experience, which is a key factor for technology adoption. Understanding whether, and how, users perceive AV behavior as human-like is therefore critical. However, existing studies do not investigate the effects of different controllers, nor do they clearly identify the visual or motion cues that shape such perceptions, leaving AV developers with limited guidance on how to design for human-likeness. This study addresses this gap by presenting an evaluation framework, based on a Turing-type experiment conducted in a moving-base driving simulator, for the perceived human-likeness of two different AV controllers. Forty participants took part in the experiment, where four different groups experienced driving trajectories generated by two different controllers and human driving styles, and were asked to classify the behaviors as either human-or AV-generated. We demonstrate our framework by evaluating a baseline model predictive controller (MPC), an advanced and widely used control technique in AVs, with a novel controller (DriViDOC) that integrates MPC with imitation learning from human driving demonstrations. Participants were significantly less able to distinguish DriViDOC from human drivers than they were with MPC, suggesting that imitation learning effectively produced more human-like behavior. In addition, we analyzed participants' justifications for their decisions, identifying key perceptual cues that contributed to their judgments, such as lane positioning and acceleration patterns, which are directly influenced by the control design. The results highlight the role of imitation learning and control tuning in shaping user perceptions of AV driving style, offering valuable insights for the development of human-like automated driving systems.
Keywords:
Artificial intelligence (AI)
Autonomous vehicles (AVs)
Turing test
Driving simulator
Journal
IF:
4.4
Papers:
3.3K
Citations:
1.3W
Organization
Cited Papers