Human Robot Interaction

Human–robot interaction (HRI) is the study and design of how people and robots work, talk, and live together.

10 frameworks, from 1945 to today · Robotics · Engineering

still a working view ended · each leads to its entry below

Overview

Human–robot interaction (HRI) is the study and design of how people and robots work, talk, and live together1. It sits at the meeting point of robotics, psychology, design, and computer science, and it asks questions that pure engineering does not: not only whether a robot can perform a task, but whether a person can understand it, trust it, stay safe near it, and want it around1. The field's history is a slow widening of that question—from keeping people out of the robot's way, to putting the robot in the person's world, to building machines that read and respond to human behavior.

Teleoperators and the First Human in the Loop

The earliest human–robot relationships were remote ones. In the 1940s, working on radioactive materials at what became Argonne National Laboratory, Raymond Goertz developed mechanical master–slave manipulators: an operator moved a handle, and a linked arm across a protective wall copied the motion23. These teleoperators kept a human in control while removing the human from danger. The pattern—human commands, machine executes, feedback returns—defined the first decades of interaction.

Teleoperation matured with the space and deep-sea programs. The 1960s and 1970s brought powered, servo-controlled manipulators and, later, force reflection, so the operator could feel what the remote arm touched. The central design problem was transparency: making the remote machine feel like an extension of the operator's own hand. This work established a durable idea—that the interface between person and machine is itself an engineering object, as important as the robot.

Safety by Separation

As industrial robots spread through factories from the 1960s onward, the dominant interaction model was avoidance. A robot arm swinging a heavy payload is dangerous, so the standard answer was to fence it off. Workcells were enclosed, interlocked, and governed by safety standards that treated the human as a hazard to be excluded during operation. This was effective and is still the backbone of industrial safety, but it made "interaction" mostly a matter of keeping the two parties apart. The unresolved problem—how to let people and robots share space without a fence—would drive the field for decades.

Supervisory Control

Full teleoperation ties up a person continuously; full autonomy was not achievable. In the 1970s Thomas Sheridan and William Ferrell at MIT proposed supervisory control, a middle ground in which the human sets goals and monitors, while the machine executes lower-level actions and the operator intervenes when needed. Their framework named the key variables—how often the human must intervene, how much feedback to give, how to allocate authority between person and machine—that HRI still reasons about. It reframed the human not as a joystick operator but as a supervisor, and it made the division of labor between people and machines a research question in its own right.

Cobots and Shared Workspace

The fence came down in the 1990s. In 1996 J. Edward Colgate and Michael Peshkin at Northwestern introduced the "cobot," a robot designed to work alongside a person, and the idea of collaborative robotics took hold: machines built to share space safely rather than be walled off4. The enabling insight was that safety could be designed into the robot itself—through limited force, rounded forms, torque sensing, and speed-and-separation monitoring—rather than enforced only by barriers. Standards for collaborative operation followed in the 2000s and 2010s5. The cobot reframed the human as a coworker, and it opened manufacturing tasks where full automation was impractical but full manual labor was costly.

The Social Turn

Through the 1990s and 2000s, a second current pushed HRI beyond the factory. Researchers began building robots meant to interact with people socially—as guides, companions, teachers, and therapeutic aids. Cynthia Breazeal's Kismet, built at MIT in the late 1990s, used expressive eyes, eyebrows, and vocal affect to engage people and to study how humans read emotion in a machine. The field's methods borrowed from psychology and human–computer interaction: controlled studies of how people respond to a robot's gaze, gesture, voice, and apparent personality. The question shifted from "can it be safe near me?" to "can it be legible, likeable, and socially appropriate?"

Humanlikeness and the Uncanny

A recurring strand asked how human a robot should look. The "uncanny valley," proposed by Masahiro Mori in 1970, held that near-human appearance can provoke discomfort before true humanlikeness is reached. Androids—robots built to resemble people closely—became both a design ambition and a test case. Work by Hiroshi Ishiguro and others produced highly humanlike and remotely operated androids and used them to probe how appearance, motion, and voice affect a person's sense of presence and acceptance. The findings were rarely clean: humanlikeness could aid engagement in some settings and raise expectations or unease in others. The lesson that took hold was that appearance must be matched to function—a machine should signal what it can actually do.

Out of the Lab

As robots moved into homes, hospitals, schools, and streets, researchers confronted a gap between short laboratory studies and long real-world use. Field and long-term deployment studies followed robots through weeks and months of daily life, tracking how people's attitudes changed, how robots broke down, and how use drifted from the designers' intentions. These studies exposed problems invisible in a single session: novelty wearing off, maintenance burden, mismatched expectations, and the difficulty of designing for people who were not researchers. They pushed HRI toward longitudinal methods and toward treating deployment as a source of knowledge rather than a final demonstration.

Trust and Transparency

If people are to rely on robots, they must calibrate their trust—neither over-relying on a fallible machine nor abandoning a capable one. From the 2000s, HRI researchers modeled trust and reliance, borrowing from automation research and human factors. A central finding was that trust should match the robot's actual reliability, and that transparency—showing what the robot is doing and why—helps people calibrate. Studies of robot behavior, explanations, and failure handling showed that how a machine communicates its limits shapes whether people use it well. This work gave the field a vocabulary for the human side of autonomy: reliance, complacency, and the design of interfaces that keep expectations honest.

Learning to Adapt

Machine learning let robots personalize. Instead of fixed behavior, systems could adapt to a particular user's preferences, pace, or style over time—adjusting a assistive arm's assistance, a tutor's difficulty, or a companion's responses. Learning-based personalization raised new interaction questions: how much adaptation is welcome, how to keep the user in control of it, and how to learn from sparse, noisy human feedback. The robot became something that changes because of the relationship, not just something that performs a task.

Conversation and Language

The most recent turn is linguistic. Early social robots used scripted speech and simple commands. As speech recognition and large language models improved through the 2010s and 2020s, robots could hold open-ended conversation, interpret natural-language instructions, and explain themselves in words. LLM-mediated interaction made language a general interface to robot behavior, but it also imported the weaknesses of language models—unreliability, opacity, and the risk of confident error—into physical systems. The open problem is grounding: connecting what a robot says to what it can safely and actually do.

Where the Field Stands

Today HRI draws on all these strands at once. Collaborative robots share factory floors under safety standards; social robots work in classrooms and care settings; teleoperation remains essential for surgery, space, and hazardous work; and language models are being wired into robot control. The field's methods run from controlled laboratory experiments to months-long field deployments, and its central questions remain open: how to build trust that is calibrated rather than blind, how to make increasingly capable and conversational machines legible and safe, and how to design interactions that hold up when the novelty fades and real people, with all their variety, are the ones in the room.

References

  1. Human–robot interaction, Wikipedia
  2. Raymond Goertz, Wikipedia
  3. Remote manipulator (History), Wikipedia
  4. Cobot (History), Wikipedia
  5. ISO 10218, Wikipedia

The frameworks of Human Robot Interaction

Each framework is a way of working the field has used: when it held, and the question it answers.

  1. 1945-Present

    Teleoperation

    How can a person control a machine at a distance?

    Fly to its tower in Nousfarer

  2. 1961-Present

    Safety by Separation

    How do we keep people safe near powerful machines?

    Fly to its tower in Nousfarer

  3. 1970-Present
  4. 1970-Present

    Supervisory Control

    How should authority be divided between person and machine?

    Fly to its tower in Nousfarer

  5. 1995-Present

    Social robot

    How can a robot engage people socially?

    Fly to its tower in Nousfarer

  6. 1996-Present

    Cobot

    How can robots and people share workspace safely?

    Fly to its tower in Nousfarer

  7. 2005-Present
  8. 2005-Present
  9. 2010-Present

    Learning-Based Personalization

    How can a robot adapt to an individual user?

    Fly to its tower in Nousfarer

  10. 2015-Present

    Language-Mediated Interaction

    How can language serve as a general robot interface?

    Fly to its tower in Nousfarer

How the ideas connect

Each arrow runs from an earlier idea to the later one it shaped. The list below says what passed between them.

TeleoperationSafety by SeparationHumanlikeness and the Uncanny ValleySupervisory ControlSocial robotCobotField and Long-Term Deployment StudiesTrust and TransparencyLearning-Based PersonalizationLanguage-Mediated Interaction

Beyond the field

Where Human Robot Interaction's ideas came from in other fields, and what they led to there: the strongest 12 of 22 links.

A city in Nousfarer: a red-carpeted stair climbing terraces to a crowned tower on an acropolis, more towers round it, forest and snowy mountains beyond.
A city in Nousfarer, the game played in the browser.

Human Robot Interaction is a planet in Nousfarer

Its city has a tower for each of these 10 frameworks. Land, walk in, and climb one concept at a time.

Land on Human Robot Interaction

Related fields

Fields its ideas are linked to

More in Robotics

See where Human Robot Interaction sits on the Map of Human Knowledge