⚡ electrical engineering

The Role of Variability in Human-Machine Interaction Experience

This study demonstrates that a human-variability-aware optimal controller significantly enhances perceived interaction quality and usability in haptic shared-control tasks, without compromising task performance, by preserving natural human movement stochasticity rather than suppressing it.

Original authors: Sean Kille, Jan Lars Hagemann, Anne Voormann, Balint Varga, Andrea Kiesel, Sören Hohmann

Published 2026-08-13

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Sean Kille, Jan Lars Hagemann, Anne Voormann, Balint Varga, Andrea Kiesel, Sören Hohmann

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are learning to ride a bike. At first, you wobble a little, your handlebars wiggle, and your path isn't a perfect straight line. That wobble isn't a mistake; it's your brain and body figuring out how to balance. Now, imagine a super-smart, invisible friend riding with you. If this friend is too bossy, they might lock your handlebars into a perfectly straight, rigid line to stop you from falling. You get to your destination faster, but you feel like a passenger in your own bike, not the rider. You lose the feeling of "I am doing this." This is the heart of a field called Human-Machine Interaction (HMI). It's the science of how humans and robots (or computers) work together, especially when they are physically touching, like holding a robotic arm or steering a car. For a long time, engineers thought the best robot was one that eliminated all human mistakes and wobbles to make things perfect. But recent thinking suggests that maybe those "wobbles"—what scientists call variability—are actually a feature, not a bug. They are the natural rhythm of how humans move. The big question is: If a robot helps us but lets us keep our natural wobbles, do we feel more in control and enjoy the ride more, even if we get to the finish line just as fast?

This paper dives into that exact question. The researchers set up a virtual game where people had to move a robotic arm from a starting point to a target point as fast and accurately as possible. They tested three different ways the robot could help (or not help). First, there was the "No Help" mode, where the person did it alone. Second, there was the "Strict Helper" mode (called lowVar), where the robot acted like a rigid guide, smoothing out every little wiggle and forcing the movement to be perfectly straight and steady. Third, there was the "Chill Helper" mode (called highVar), which used a special new control strategy. This robot was smart enough to help with the hard parts (getting to the target) but deliberately let the person keep their natural, slightly wobbly movements in the directions that didn't matter for the goal.

The results were pretty cool. When they looked at how well the people did the task, the "Strict Helper" and the "Chill Helper" were basically tied. Both got the job done just as well, and both were much better than doing it alone. The "Chill Helper" didn't make people slower or less accurate. However, when they asked the people how the experience felt, the "Chill Helper" won hands down. People rated the "Chill Helper" as much more usable and easier to use than the "Strict Helper." It turns out that letting the robot respect your natural, slightly messy movement style made the whole interaction feel more intuitive and comfortable, even though the robot was still doing the heavy lifting to ensure success.

The study didn't find that the "Chill Helper" made people feel more "in charge" (a feeling called the sense of agency) or more confident in their skills compared to the "Strict Helper." In fact, both helpful modes made people feel more capable and in control than doing it alone, likely because they were succeeding at the task. But the specific feeling of "this system is easy to use" was significantly higher when the robot allowed for natural variability. The researchers suggest that by not over-correcting the human's natural movements, the robot feels less like a bossy dictator and more like a helpful partner. This study proves that you don't have to sacrifice performance to get a better experience; in fact, by respecting the natural "wobble" of human movement, you can make robots feel friendlier and more human-like without making the job harder.

Technical Summary: The Role of Variability in Human–Machine Interaction Experience

Problem Statement

Physical human–machine interaction (pHMI) is increasingly shifting toward tightly integrated collaboration where humans and machines exchange physical forces to achieve shared tasks. While conventional shared-control and haptic-assistance strategies effectively optimize objective task performance, they typically rely on deterministic human models that ignore the stochastic nature of human motor behavior. Specifically, these systems often suppress natural movement variability, treating it as noise to be minimized rather than a structured feature of motor control.

The paper argues that systematically constraining task-irrelevant variability may alter the sensorimotor contingencies through which users generate predictions and attribute action outcomes. This mismatch between the controller's internal model and the user's natural movement statistics could degrade the user's sense of agency, perceived usability, and overall interaction experience, even if task performance remains high. Despite the established role of structured variability in motor control, its specific impact on the human interaction experience in tightly coupled pHMI has not been empirically explored.

Methodology

Experimental Design

The study employed a within-subject design with 41 participants performing a physically coupled haptic point-to-point task. The independent variable was the interaction mode, consisting of three conditions:

noSup (No Support): A baseline condition with no automation, capturing natural human movement patterns.
lowVar (Variability-Constraining): A conventional controller mode designed to minimize overall movement variability.
highVar (Variability-Respecting): A novel Human-Variability-Respecting Optimal Control (HVROC) mode designed to preserve natural human movement variability patterns in task-irrelevant dimensions while maintaining task accuracy.

System Implementation

Hardware: A KUKA LBR iiwa 14 R820 robotic arm equipped with a force-torque sensor served as the haptic interface.
Simulation: The robot was virtually coupled to a simulated linear point-mass system (mass-damper dynamics) in a 2D workspace.
Control Logic: The automation applied additive force inputs based on a state-feedback control law.
- The lowVar mode used a stationary HVROC solution with high feedback gains ( $L_p = L_d = 150$ ) to aggressively dampen deviations.
- The highVar mode employed position-error-dependent gain scheduling. It utilized lower initial gains ( $L_p=75, L_d=20$ ) to allow natural variability during the movement phase, increasing gains linearly only as the system approached the target ( $e < 1/3 e_{start}$ ) to ensure stable deceleration.
Task: Participants performed 82 movements across four directions, balancing speed and accuracy. Automation was activated only when human-applied force exceeded 5 N to isolate the effects of the control strategy from reaction time delays.

Measures

Behavioral Metrics:
- Task-Irrelevant Variability: Measured as the maximum positional variance ( $V_{max}^y$ ) orthogonal to the movement direction during the mid-movement phase.
- Task Performance: Measured via settling time, endpoint error, and endpoint variance in the target region.
Interaction Experience Metrics:
- Sense of Agency: Perceived control over actions and outcomes.
- Self-Efficacy: Belief in one's capability to perform the task.
- Flow: State of focused engagement.
- Usability: Assessed via the QUEAD questionnaire, including subscales for perceived usefulness, ease of use, and emotional response.

Key Contributions

Empirical Investigation of Variability Structure: The study provides the first empirical evidence regarding the causal relationship between task-irrelevant movement variability and human interaction experience in tightly coupled pHMI.
HVROC Implementation: The paper demonstrates the feasibility of a variability-aware optimal controller that explicitly manipulates task-irrelevant variability without compromising task-relevant performance.
Decoupling Performance and Experience: The research establishes that control strategies can be functionally equivalent in terms of objective task success while being experientially distinct.

Results

Task Performance

Variability Manipulation: The highVar mode successfully preserved significantly more task-irrelevant variability ( $M=0.65 \times 10^{-4} m^2$ ) compared to the lowVar mode ( $M=0.33 \times 10^{-4} m^2$ ), though both were lower than the noSup baseline ( $M=1.11 \times 10^{-4} m^2$ ).
Task Success: Both assisted modes (lowVar and highVar) significantly outperformed the noSup baseline in settling time, endpoint error, and endpoint variance.
Equivalence: Crucially, there were no significant differences in task-relevant performance (settling time, endpoint error, or endpoint variance) between the highVar and lowVar modes. This confirms that preserving natural variability patterns does not come at the cost of objective task performance.

Human Interaction Experience

General Experience: Both assisted modes yielded significantly higher ratings for sense of agency, self-efficacy, flow, and usability compared to the noSup baseline.
Variability Impact: While agency, self-efficacy, and flow did not differ significantly between the two assisted modes, usability was significantly higher in the highVar condition compared to lowVar.
Subscale Analysis: The improvement in usability for highVar was primarily driven by significantly higher ratings in perceived ease of use and a positive trend in emotional response. Perceived usefulness, which correlated with task success, showed no significant difference between the assisted modes.

Significance and Claims

The paper claims that preserving structured, task-irrelevant movement variability is a critical design principle for human-centered pHMI. The findings suggest that:

Experience is Distinct from Performance: Control policies can be experientially distinct even when they produce identical objective task outcomes.
Usability Enhancement: Variability-aware control enhances perceived usability, specifically the "ease of use" and emotional quality of the interaction, likely by aligning the system's dynamics with the user's internal motor models and natural sensorimotor predictions.
Design Implications: For tightly coupled systems, especially in domains like healthcare and rehabilitation where long-term acceptance is vital, control strategies should not merely minimize error but should respect the statistical structure of human movement. This approach allows for strong task support while preserving the user's active, self-attributed role in the collaboration.

The authors conclude that variability-aware control should be regarded not as an optional refinement, but as a deliberate design principle to achieve both effectiveness and sustained interaction quality in human–machine systems.