Linking Behaviour and Perception to Evaluate Meaningful Human Control over Partially Automated Driving

Partial driving automation creates a tension where drivers retain legal responsibility despite significantly reduced active control, necessitating meaningful human control (MHC) as an evaluative framework. This study investigated how different interaction modes—haptic shared control (HSC) and traded control (TC)—affect driver perceptions of MHC by linking behavioral metrics to subjective scores.

How it works

The research employed a mixed-methods approach, integrating detailed driving interaction data with subjective perception scores from post-trial surveys and qualitative feedback. This methodology was specifically designed to address the gap in the literature where existing evaluations of MHC are often indirect, combining behavioral metrics derived from telemetry data with questionnaires explicitly designed to assess MHC based on established properties (e.g., tracking and tracing conditions).

Experimental Setup and Control Modes

Twenty-four participants completed a simulator study involving silent automation failures under two modes: haptic shared control (HSC) and traded control (TC). In the driving simulator, participants interacted with an automated vehicle through a haptic steering wheel. HSC involves both the driver and automation applying torques simultaneously, supporting smoother collaboration, whereas TC relies on explicit handovers where authority is explicitly transferred based on torque thresholds. The experiment included baseline fully manual control trials alongside nine automated driving trials under either HSC or TC, simulating a silent automation failure to test critical intervention scenarios.

Key Behavioral Metrics and Hypotheses

The study formulated trial-level behavioral metrics to quantify interaction dynamics, including:

1. Reaction time (time between automation initiation and first human response).

2. Average conflict in steering torques (negative signed product of human and automation torques exceeding a threshold).

3. Maximum steering torque (effort exerted by the participant).

4. Steering reversal rate (number of reversals per second).

5. Jerk of vehicle trajectories (smoothness quantified as root-mean-squared jerk).

6. Time to collision (TTC) during failure events.

The hypotheses linked these metrics to subjective perception scores derived from Likert-scale questions assessing MHC properties such as 'sufficient control' and 'AV understood me'. For instance, the study hypothesized that reaction times would be negatively correlated with subjective scores for item A1 (‘sufficient control’) and that conflict is negatively correlated with subjective scores A2, A3, and A4 (‘AV understood me’, ‘I understood AV’, and ‘working together’, respectively).

Confirmatory Findings on Control Modes

The confirmatory analysis using Linear Mixed-Effect Models (LMMs) revealed significant differences between control modes for several metrics. For example, reaction times in simulated automation failures were significantly lower for HSC than for TC (β = 1.33, p < 0.001). Furthermore, maximum steering torque was lower for HSC than for TC (β = 0.57, p < 0.001). The analysis also indicated that the number of takeovers was larger for HSC than for TC (β = -1.37, p < 0.001).

Qualitative Insights and Design Implications

The qualitative analysis categorized feedback into three themes: attitudinal factors, interaction factors, and trajectory factors. Participants frequently cited mismatches in intentions between the driver and automation, lack of safety, and resistance to driver inputs as contributing to the reduction of perceived MHC. The exploratory analysis suggested that HSC generally supported better perceptions for 'sufficient control', 'AV understood me', and 'working together' compared to TC. The study concluded that future designs should prioritize effortless driver interventions, transparent communication of automation intent, and context-sensitive authority allocation.

Conclusion on Meaningful Human Control

The study demonstrated that HSC yielded a more favorable MHC profile than TC across key subjective perceptions. Specifically, HSC was associated with higher scores for 'sufficient control', 'AV understood me', and 'working together'. While responsibility perceptions were less dependent on the control mode, the overall findings underscore that meaningful human control emerges from how human-automation interaction sustains engagement and preserves a sense of authorship over vehicle behavior. Guidelines suggest ensuring interventions are smooth, improving mutual communication of intent, and validating design choices through structured experiments.

The gist: HSC supports meaningful human control better than TC across perceptions of 'sufficient control', 'AV understood me', and 'working together'.

Based on the scientific paper, here are specific improvements that can be made to AI systems, categorized by how they align with Meaningful Human Control (MHC):

)Meaningful Human Control (MHC) Enhancements for Partially Automated Driving AI Systems:

1. Enhanced Intent Communication and Transparency Module:

2. Dynamic Authority Allocation Engine:

3. Perception-Aware Risk Signaling System:

4. Adaptive Interaction Parameter Tuning (for HSC/TC):

)Specific Improvements and Capabilities of the Improved AI System:

1. The system should implement a module that ensures transparent communication of its intended maneuvers and future actions to the driver (addressing MHC Property 2).

2. The system must feature a dynamic authority allocation engine capable of smoothly transitioning control authority based on real-time driver intent and situational context, avoiding abrupt transitions or sudden disengagement (improving HSC/TC interaction smoothness).

3. The AI should incorporate a perception-aware risk signaling system that provides subtle haptic guidance aligned with the driver's intent, rather than relying solely on forceful torque resistance (addressing the finding that subtle haptic guidance had a positive effect).

4. The control system should be designed to minimize conflict in steering torques between the human and automation, ensuring that when conflict occurs, it is managed in a way that aligns with driver understanding rather than creating feelings of being misunderstood or taken over (addressing the negative correlation between conflict and 'AV understood me').

)What the Improved AI System Can Do:

This improved AI system will transition from a purely task-oriented controller to a collaborative partner that actively fosters human agency. Specifically, it can:

1. Perform complex maneuvers (like overtaking or lane changes) while continuously communicating its intentions through clear, non-intrusive cues (e.g., subtle torque feedback).

2. Intelligently interpret driver inputs and intentions in real-time to modulate its own authority level seamlessly, allowing the driver to feel like they are working together rather than being subjected to rigid pre-set modes (HSC optimization).

3. Proactively signal potential safety risks using contextually relevant, non-threatening cues that align with the driver's current state, thereby reducing driver complacency and maintaining high engagement.

4. Ensure that when a handover or override is necessary, it is executed in a manner that maximizes the perception of shared control and responsibility, ensuring the driver feels capable of contributing to the outcome even during system-led maneuvers (improving perceived 'Sufficient Control' and 'Working Together').

Partial driving automation creates a tension: drivers remain legally responsible while being less active in control. Meaningful human control (MHC), a normative framework that can potentially address this tension, proposes that automated systems are designed to track relevant human reasons and that humans should at all times remain in control and be responsible. However, empirical methods for evaluating whether systems are under MHC remain underdeveloped. In this driving simulator study, we investigated the extent to which 24 drivers experienced MHC when interacting with partially automated driving systems under two modes - haptic shared control and traded control. During overtaking manoeuvres on a two-lane, two-way road with fully automated longitudinal control and partially automated lateral control, drivers' actions were necessary to prevent crashes due to silent automation failures. Starting from hypotheses derived from the properties of systems under MHC, we used a mixed-methods approach that links behavioural metrics, subjective post-trial ratings, and qualitative feedback to assess drivers' perception of responsibility and control. A confirmatory analysis indicated a negative correlation between the perception of the automated vehicle understanding the driver and conflict in steering torques. Qualitative feedback revealed that mismatches in intentions between the driver and automation, lack of safety, and resistance to driver inputs reduced perceived MHC, while subtle haptic guidance aligned with driver intent had a positive effect. Thus, future designs should prioritise effortless driver interventions, transparent communication of automation intent through haptic, visual or auditory cues, and clear authority allocation to strengthen meaningful human control in partially automated driving.