Agency – Control

Control is a central part of agency in human-robot interaction. It concerns whether a robot stays within the user's physical and practical boundaries, or whether it begins to act in ways that override, constrain, or pressure the user. In everyday settings, this can be subtle: a rehabilitation robot that pushes a patient beyond their comfort level, a care robot that insists on an action, or a domestic assistant that limits what a user can choose to do. The concern is not only about extreme cases of coercion, but also about small, repeated moments where the user's ability to decide and act is weakened. As robots become more present in sensitive domains such as healthcare, education, and home assistance, maintaining meaningful user control becomes a defining ethical requirement.

In our framing of agency, control is understood as the requirement that robots should not physically constrain or force users into actions they do not wish to take. This position emerged both from the literature and from focus group discussions, where participants repeatedly emphasised the importance of being able to override or decline a robot's suggestions or actions. Earlier work often placed this concern under "empowerment," but it is more precisely captured by agency, since it directly relates to a user's ability to act according to their own intentions. The appropriate level of control is also highly context-dependent. For instance, in rehabilitation scenarios, a robot may need to guide or encourage movement, but it should still adapt to the user's pace and remain responsive rather than imposing a fixed or rigid trajectory. The core idea is that assistance should not quietly become constraint.

The literature reflects this tension between assistance and overreach across many domains. In discussions of ethical design, there is a consistent emphasis on keeping humans "in the loop" for decision-making, ensuring that responsibility and control do not fully shift to the robot or system. In healthcare-related rule selection studies, participants rejected forms of physical restraint outright and emphasised that robots should never constrain a person, even when aiming to help. At the same time, there is nuance: repeating prompts or actions, such as reminders for medication, was seen as acceptable and not necessarily a violation of autonomy. This indicates that control is not about removing all robot initiative, but about carefully distinguishing between guidance and coercion. Concerns about loss of control also appear in broader reflections on care and aging. Some older adults explicitly emphasized the importance of retaining the right to decline a robot's suggestions, particularly in contexts like physical exercise. Others expressed discomfort with the idea of machines overriding human wishes, even when framed as assistance. More extreme concerns highlight the risk of "loss of freedom," where robot-mediated care could unintentionally restrict a person's life to the point of feeling confined within their own home. In more everyday terms, participants have described worries about robots becoming "bossy," issuing instructions in a way that feels directive rather than supportive. Even when framed as beneficial, such as improving safety, there is ongoing debate about whether increased safety justifies reducing user choice or effectively pressuring users into robotic assistance. Together, these perspectives show that control is not a peripheral issue in HRI, but a recurring concern that shapes how acceptable and trustworthy robotic systems feel in practice.