Research

Human-Centered Analysis and Behavior Modeling for Safe Remote Driving

Funded by the National Science Foundation (NSF)

Remote Driving Human Behavior End-to-End Latency

Remote driving is emerging as a critical safeguard for automated transportation systems, enabling human operators to intervene when autonomy reaches its limits. This research examines how end-to-end latency shapes human behavior, control performance, and safety, with the goal of defining human-centered solutions for reliable and scalable remote vehicle operation.

Remote driving is a mode of vehicle operation in which a human operator remotely performs the dynamic driving task (DDT) when an automated driving system (ADS) encounters situations beyond its operational design domain (ODD). As part of remote assistance frameworks, it enables operators to safely guide vehicles until automated driving can resume, supporting the safe deployment of automated transportation systems.

Our studies show that delay does not simply make remote driving harder in a gradual way. Instead, control quality can remain relatively stable and then deteriorate quickly once latency approaches a critical range, making it essential to understand where remote driving shifts from manageable to unstable.

Summary

Research question. Can humans reliably control remotely driven vehicles under delayed feedback, and if so, who can adapt and how?

Why it matters. These findings show that remote driving is not just an engineering problem. It is a coupled system of human behavior, system latency, and cognitive adaptation. Designing safer remote driving therefore means identifying who can adapt, supporting how adaptation happens, and avoiding the hidden critical breakpoint where control becomes unstable.

Related journal publications

X. Ma, A. D. McDonald. Modeling Driver Adaptation Patterns During Remote Driving Under Latency. Under review (Human Factors), 2026.

  • Shows that adaptation is not a smooth learning curve. Drivers move between distinct control strategies rather than gradually getting better.
  • Identifies a transitional state as the trial-and-error window where drivers actively test and revise their control strategies under delay.
  • Finds that individual differences are decisive: some drivers adapt rapidly, some gradually, and some fail even under comparatively low latency.

X. Ma, A. D. McDonald. Identifying Critical Breakpoints and Latency Effects During Remote Backup Driving. IEEE Transactions on Human-Machine Systems, 2026.

  • Reveals an invisible boundary, a critical breakpoint, around 350 to 480 ms where trust drops, control destabilizes, and performance deteriorates sharply.
  • Shows that remote-driving risk is context-dependent rather than uniform: curved lane keeping is most fragile, straight driving is relatively stable, and lane changes are more robust than expected.
  • Shows that trust is not simply given by system performance; it emerges from perceived controllability, rising when drivers stabilize control and falling when control breaks down.

Dynamic Trust and Reliance in Malfunctioning Driving Automation and Risky Environments

Funded by the National Science Foundation (NSF)

Driving Automation Trust Calibration Reliance

As driving automation becomes increasingly common, safety depends not only on system capability but also on how drivers adapt their trust and reliance when failures or risks emerge. This research examines the dynamic interplay between automation malfunctions, environmental hazards, and human decision-making to develop strategies that calibrate trust, prevent misuse or disuse, and improve human–automation safety.

This work studies how drivers form, update, and act on trust while interacting with conditional driving automation. The core question is whether reliance matches system capability, especially when automation performance varies and drivers must decide when to intervene.

Rather than treating trust as a fixed attitude, this research examines how reliance shifts over time as automation malfunctions, environmental risk, and driver uncertainty unfold together in real driving decisions.

Summary

Research question. How do drivers calibrate trust and reliance when automated vehicle performance changes across situations and risks emerge?

Why it matters. Safer driving automation depends on more than improving the system itself. It also requires understanding when drivers over-rely, under-rely, or intervene too late, so that interfaces, adaptive support, and policy can better align human reliance with true system capability.

Related journal publications

X. Xiao, X. Ma, A. D. McDonald, R. K. Mehta. What leads to reliance on automated vehicles? An inferential analysis of responses to variable AV performance. Applied Ergonomics, 2025.

  • Reveals reliance inertia as the hidden engine of AV reliance: once drivers choose to rely, they are far more likely to keep relying, even after automation failures or degraded performance.
  • Shows that reliance is scenario-dependent rather than uniform: each driving context activates a different mix of trust, situation awareness, workload, demographics, and prior reliance history.
  • Reframes trust as only part of the reliance story: situational trust matters more than dispositional trust in most cases, but actual reliance is shaped more strongly by behavioral momentum than by trust alone.