Loading field map
Human Robot Interaction (HRI) is the field of study dedicated to understanding, designing, and evaluating robotic systems for use by or with humans. It is a fundamentally interdisciplinary endeavor, drawing on robotics, computer science, human factors engineering, cognitive psychology, sociology, and design. At its core, HRI is not merely the study of how people use machines; it is the study of a new class of social and physical partners. The central questions concern how humans perceive, trust, coordinate with, and are affected by autonomous and semi-autonomous machines, and conversely, how robots should be designed to communicate, act, and adapt in ways that are legible, predictable, and safe for the people around them.
The field's stakes are high because robots are increasingly leaving the confines of factory cages. They are being deployed in homes, hospitals, schools, and public spaces, where they must operate in unstructured environments alongside people who are not trained engineers. The success of these deployments hinges not just on the robot's mechanical or algorithmic competence, but on its ability to interact with humans in a manner that is socially appropriate and psychologically acceptable. HRI, therefore, is the discipline that bridges the gap between the robot as a tool and the robot as a teammate.
A foundational distinction in HRI is between the problem of the robot and the problem of the human. The former is a classical engineering problem: how to make a machine perceive, plan, and act. The latter is the central concern of HRI: humans are not deterministic, fully observable, or rational agents. They have emotions, biases, and fluctuating attention. They make mistakes, and they anthropomorphize—attributing intentions, personality, and even moral standing to machines that do not possess them. HRI research is therefore as much about modeling and predicting human behavior as it is about controlling robot behavior.
This focus on the human gives rise to a set of enduring questions. How does a person form a mental model of a robot's capabilities and limitations? How does trust develop, erode, and get repaired in human-robot teams? How does a robot's appearance—from a mechanical arm to a humanoid face—shape expectations and performance? How should a robot signal its next action so that a nearby human can anticipate it? And how should a robot respond when a human's command is ambiguous, unsafe, or socially inappropriate? These questions are not peripheral; they are the defining problems of the field.
The intellectual roots of HRI lie in two distinct traditions that converged in the late twentieth century. The first is teleoperation, the direct manual control of a remote machine, which emerged in the mid-twentieth century for handling radioactive materials and, later, for underwater and space exploration. Teleoperation forced engineers to confront human factors: how to present remote sensor data, how to design control interfaces, and how to manage time delays. The second root is artificial intelligence, which from its inception in the 1950s and 1960s imagined machines that could converse, reason, and assist people. Early AI research on natural language understanding and problem-solving laid the groundwork for the idea of a robot as a communicative agent, not just a remotely operated tool.
The modern field of HRI began to coalesce in the 1990s and early 2000s, driven by the proliferation of affordable, safe, and increasingly autonomous robots. The rise of social robotics—robots designed explicitly to engage people socially, such as the robotic pet AIBO or the humanoid ASIMO—created a new research agenda. Researchers began to ask not just "Can a robot do this task?" but "How should a robot behave so that a person wants to work with it?" The first dedicated HRI conferences and journals appeared in the early 2000s, marking the field's institutionalization. Since then, HRI has grown rapidly, absorbing methods from psychology (experimental design, survey instruments) and anthropology (ethnographic observation) to study human-robot encounters in the wild.
The field is not organized into a single dominant paradigm but is instead structured by several overlapping research traditions, each with its own assumptions, methods, and objects of study. These approaches are best understood as complementary lenses rather than rival schools.
This approach treats the human-robot system as a joint cognitive system to be optimized. Its methods are borrowed from experimental psychology and human factors engineering: controlled laboratory studies, reaction-time measurements, eye-tracking, and standardized questionnaires. The goal is to measure and predict human performance, workload, and situation awareness when interacting with a robot. A classic problem is the "mode confusion" that arises when a human operator misjudges what a semi-autonomous robot is currently doing or will do next. The solution is typically better interface design: clearer displays, more intuitive control schemes, and predictable automation behavior.
This tradition is strongly empirical and quantitative. It has produced robust findings about, for example, the "out-of-the-loop" performance problem, where humans supervising an autonomous system fail to notice errors or to retake control effectively. Its limitation is that it often treats the human as an information processor, potentially underplaying the emotional, social, and cultural dimensions of interaction. Nevertheless, its methods remain the gold standard for evaluating any HRI system.
In contrast, the social robotics tradition centers on the robot as a social actor. Researchers in this vein deliberately design robots with faces, voices, gestures, and personalities, drawing on social psychology and communication theory. The central assumption is that humans respond to robots with the same social heuristics they use with other humans, and that this response can be harnessed for better interaction. A robot that apologizes after a mistake, for example, may be trusted more than one that simply beeps an error code.
This approach has produced influential findings on the "uncanny valley"—the hypothesis that a robot or animated character that looks almost, but not perfectly, human elicits feelings of revulsion—and on the power of social presence to increase engagement and compliance. Its methods are often experimental but may also include long-term field studies in homes or schools. Its limitation is that the social effects it studies can be fragile, context-dependent, and ethically fraught. A robot that is too persuasive, too cute, or too human-like can mislead users about its actual capabilities, raising concerns about deception and over-reliance.
A third tradition, more recent and more critical, argues that HRI has historically been too robot-centered, focusing on what robots can do rather than what people actually need or want. This approach, sometimes called human-centered robotics, draws on participatory design, value-sensitive design, and science and technology studies. It insists that the questions of HRI cannot be separated from their social, cultural, and political context. Who benefits from a care robot? Whose labor does it replace? What values are embedded in its algorithms?
This tradition is more likely to use qualitative methods: interviews, focus groups, ethnographic observation, and co-design workshops where potential users are involved in the design process from the start. It has been influential in shaping discussions about robot ethics, privacy, and the deployment of robots in care settings. Its limitation is that it can be less prescriptive, offering critiques and design principles rather than testable hypotheses. However, it has become an essential corrective to the field's early optimism, reminding researchers that a robot that works in a laboratory may fail—or cause harm—in a real community.
A fourth approach, rooted in machine learning and control theory, treats the interaction itself as a data source. Here, the robot is not pre-programmed with social rules but instead learns from human feedback, demonstration, or its own trial-and-error experience. The goal is to create robots that personalize their behavior to individual users, adapting their pace, communication style, or task strategy over time. Methods include reinforcement learning from human rewards, inverse reinforcement learning (inferring a human's goal from their actions), and learning from demonstration.
This approach is powerful because it promises robots that can handle the infinite variability of human behavior without being explicitly programmed for every case. Its limitation is that learning from humans is noisy, slow, and potentially unsafe. A robot that learns by trial and error in a home environment might make dangerous mistakes before it converges on good behavior. Moreover, learned behavior can be opaque, making it difficult for a human to understand why the robot is acting as it does—a problem that directly undermines the legibility that other HRI traditions seek to create.
The present landscape of HRI is characterized by a productive, if sometimes uneasy, coexistence of these approaches. A typical research project might combine a human-factors evaluation of a new interface, a social-robotics design for the robot's expressive behavior, a participatory design workshop to gather stakeholder input, and a learning algorithm that adapts the robot's behavior to the user. The field's conferences and journals are notably pluralistic, publishing both controlled experiments and ethnographic studies, both algorithmic contributions and design critiques.
Several durable themes cut across these approaches. Trust remains a central concept, studied both as a psychological state (measured by questionnaires) and as a behavioral pattern (measured by whether a human intervenes in a robot's actions). Transparency and explainability have become pressing concerns as robots make more autonomous decisions; a robot that cannot explain its actions is unlikely to be trusted or safely supervised. Long-term interaction is another growing focus, as researchers move beyond single-session studies to understand how relationships with robots change over weeks or months. Finally, safety and ethics are no longer afterthoughts but core design constraints, encompassing physical safety (a robot must not harm a person) and psychological safety (a robot must not manipulate, deceive, or distress a person).
The field's greatest challenge is also its defining feature: the irreducible complexity of the human. Unlike a purely technical system, a human-robot system cannot be fully specified in advance. People are unpredictable, culturally varied, and emotionally responsive. HRI has therefore developed as a discipline that is comfortable with this uncertainty, combining the rigor of engineering with the interpretive sensitivity of the social sciences. Its future lies not in resolving this tension but in managing it—designing robots that are simultaneously competent tools and acceptable social partners, and understanding the humans who must live with them.