AI Glossary / Reinforcement learning and robotics / Intermediate

Human - robot interaction

Definition

Human - robot interaction, or HRI, is the study and design of how people and robots communicate, influence one another, and work or share environments together. It includes remote operation, nearby physical interaction, and social exchanges. The field combines robotics with human factors, psychology, design, and other disciplines. Its concern is the complete interaction, rather than robot performance considered independently of the people involved.

What it means

A robot's usefulness depends partly on whether people can understand and direct it. A clear signal that a robot is waiting may prevent confusion; an ambiguous movement may create uncertainty even when the movement is mechanically correct.
People involved in HRI are not limited to trained operators. They can include collaborators, customers, bystanders, maintainers, and others affected by the system. Their goals, abilities, expectations, and access to information can differ. Designing for one group does not automatically produce a suitable interaction for another.

How it works

Define the setting, participants, tasks, and relationship between human and robot. Remote interaction may involve a person operating a robot elsewhere, while proximate interaction involves people and robots in the same environment. Physical manipulation, mobility, and social activity create different design requirements.
Specify what information must pass between participants and through which channels. Controls, displays, speech, gestures, lights, sounds, and movement can all communicate. Choose channels that fit the task and provide understandable feedback about success, uncertainty, or failure.
Decide how control and responsibility are shared. People need appropriate ways to intervene, correct errors, or request help. A robot's inferred interpretation of a human action should not automatically be treated as certain, especially when alternatives would lead to different outcomes.
Evaluate with representative people and tasks. Combine objective measures such as completion, errors, and intervention time with carefully designed assessments of workload, usability, and trust. Observe unexpected behavior and accessibility barriers. A short demonstration with developers is not a substitute for evidence about the intended users and operating context.

Three practical examples

• A remote inspection interface shows an operator the robot's current status, communication delay, and camera view so commands can be interpreted in context.
• A service robot signals when it is waiting for a person to collect an item, reducing uncertainty about the next action.
• A shared workstation provides clear ways for a worker to pause the robot, resolve an exception, and resume the task.

Think of it this way

Think of HRI as designing the conversation around action. The conversation might use buttons and motion instead of spoken words, but both sides still need to understand enough about what is happening. A robot that executes a command correctly can still be difficult to use if the person cannot tell whether it heard, understood, or completed the request.

Worked example

Suppose an invented usability study compares two interfaces for requesting a robot-delivered item. Each interface is tested on 50 requests under comparable conditions. Interface A completes 42 requests without human recovery; interface B completes 46.
The observed completion rates are 42 / 50 = 84% and 46 / 50 = 92%. The difference is 8 percentage points. Relative to A, the increase is (92 − 84) / 84 ≈ 9.52%, so the percentage-point difference and relative percentage increase should not be confused.
Assume each successful request takes 30 seconds and each unsuccessful request consumes 60 seconds before the trial ends. For A, total time is 42 × 30 + 8 × 60 = 1,740 seconds, averaging 34.8 seconds per request. For B, it is 46 × 30 + 4 × 60 = 1,620 seconds, averaging 32.4 seconds.
These calculations include failed attempts in the time accounting. Looking only at successful requests would show identical 30-second times and miss an important difference. However, the small illustrative dataset does not establish statistical significance, causality, accessibility, or suitability for deployment. Study design, participant differences, uncertainty, and the consequences of failure still require examination.

Illustrative case study

In this fictional case, Larch Visitor Centre trials a robot that directs guests to exhibits. Developers initially judge success by whether the robot displays the correct route after a request.
Observation reveals that some visitors cannot tell when the robot is listening, while others assume it has understood before a request is complete. The team revises turn-taking signals and adds a clear correction option. It tests with visitors who differ in familiarity with the system and preferred interaction methods.
The evaluation records completed requests, misunderstandings, recovery steps, and visitor feedback. The team treats social appeal as one aspect of the experience, rather than a replacement for understandable and usable behavior.

Common misconceptions and mistakes

HRI is not limited to humanoid robots or speech. A remotely operated machine and a mobile delivery platform can raise important interaction questions without resembling a person.
Another mistake is assuming that a friendly appearance produces appropriate trust. People may overestimate capability, misunderstand autonomy, or struggle to notice failures. The design should support informed expectations.
Satisfaction alone does not prove task effectiveness, and successful task completion alone does not establish low workload or accessibility. Evaluation should match the actual participants and context instead of relying on one convenient metric.

Business use

Businesses benefit from HRI work when robot adoption depends on understandable operation, reliable handovers, effective recovery, and appropriate expectations. Interaction design can influence training effort, support needs, acceptance, and operational errors.
Include affected people early, define meaningful outcomes, and observe actual use. Track whether improvements work across user groups and over time. Make responsibilities, system limitations, and intervention routes clear within the practical workflow.

People also ask

Is HRI the same as human - robot collaboration?

No. Collaboration concerns shared tasks; HRI also includes other forms of interaction.

Does HRI require artificial intelligence?

No. Interaction design matters for robots with different levels of automation.

Can remote operation be HRI?

Yes. Spatial separation changes communication and control requirements but does not remove the interaction.

Check your understanding

Why are the example's rates different by 8 percentage points?

Answer: They are 92% and 84%.

Why include unsuccessful requests in time measurements?

Answer: They consume user and system time and affect the experience.

Why study more than satisfaction?

Answer: Usability, effectiveness, workload, accessibility, and understanding can diverge.

Related terms

• Human - robot collaboration
• Teleoperation
• Usability
• human factors
• shared autonomy
• social robotics
• calibrated trust
• multimodal interaction

Learn next

Learn next how to design an HRI study with representative participants and tasks. Compare objective performance measures with workload and usability evidence, then study how communication delay and shared control change interaction requirements.

• Usability

More in Reinforcement learning and robotics

• Greedy policy - A greedy policy chooses an action with the highest current estimated value at each decision point.
• GTD2 - GTD2 is a gradient temporal-difference algorithm for learning a linear value-function approximation using a primary parameter vector and an auxiliary vector.
• Hierarchical reinforcement learning - Hierarchical reinforcement learning is an approach to learning sequential decisions at more than one level of abstraction.
• Intrinsic curiosity module - An intrinsic curiosity module, or ICM, is a reinforcement-learning component that generates an internal reward from errors in predicting the consequences of...
• Inverse reinforcement learning - Inverse reinforcement learning is the task of inferring a reward or cost function from observed behavior in a sequential decision-making setting.
• Keepout - A keepout is a designated area that autonomous robot navigation should avoid or must not enter.

Sources

Goodrich and Schultz, Human-Robot Interaction survey, Introduction: https://humanrobotinteraction.org/1-introduction/ Goodrich and Schultz, What Defines an HRI Problem?: https://humanrobotinteraction.org/4-what-defines-an-hri-problem/ NIST, Performance of Human-Robot Interaction: https://www.nist.gov/programs-projects/performance-human-robot-interaction

Previous: Human - robot collaboration Next: Human system integration