Wizard of Oz

• Also called: WoZ, Wizard-of-Oz prototyping
• Category: Control › Human in the loop
• Tags: human-robot-interaction, user-studies, research-methods, teleoperation, social-robotics

A study method in which a hidden human operator, the wizard, controls some or all of a robot's behavior while participants interact with the robot as if it were autonomous.

Definition

In a Wizard of Oz (WoZ) study, a person out of the participant's view, the wizard, operates part or all of a robot's behavior, such as its speech, gestures, navigation or choices, while the participant interacts with it as if it were acting on its own. The name refers to the hidden operator behind the curtain in The Wizard of Oz. The method came from human–computer interaction research on natural-language interfaces, where it let designers study how people would talk to systems that could not yet be built, and it became a standard method in human–robot interaction (HRI).

WoZ lets researchers study how people respond to capabilities that do not yet work reliably, such as open conversation, social perception or fluent turn-taking, and test design ideas before the underlying technology exists. Wizard control ranges from full teleoperation to choosing among prepared behaviors, and many studies mix wizarded components with autonomous ones, for example autonomous navigation with wizarded dialogue. Good practice includes a written wizard protocol, training and consistency checks for wizards, logging of every wizard action, and clear reporting of which behaviors were wizarded.

The method has known limits. Results describe how people respond to an idealized robot rather than to the system that can actually be built, and the wizard's judgment and reaction time become part of the experimental treatment. Participants may suspect the deception, which can change how they behave. Because participants are deliberately misled about the robot's autonomy, ethics review and a debriefing after the session are standard. Logs of wizard actions are also used afterward to design autonomous behavior or as demonstrations for learning, which makes them useful beyond the original study.

Why it matters

Many HRI questions concern how people react to robot abilities that are not yet reliable enough to deploy. Wizard of Oz makes those studies possible, but it also limits what their findings mean, so reading HRI results well requires knowing which parts of the robot were controlled by a person.

Key applications

• Prototyping conversational and social robot behavior before it can be automated
• Studying trust, engagement and acceptance in human–robot interaction
• Long-term and in-the-wild studies where autonomy is incomplete
• Collecting interaction data to design or train autonomous behaviors
• Early testing of assistive and service robot concepts with intended users

In practice

An HRI team wants to know whether a social robot that changes its role during a group game keeps players engaged over several sessions. Reliable recognition of each player's state, and a sound policy for choosing roles, are not yet available. A wizard in an adjacent room therefore watches the camera feeds and selects the robot's role and utterances from a prepared interface, while speech synthesis and gestures run automatically. The wizard follows a written decision protocol, and every selection is logged with a timestamp. After the sessions, participants are debriefed and asked whether they believed the robot was autonomous, and the team reports which behaviors were wizarded alongside the results.

Common misconceptions

• Wizard of Oz just means the robot is teleoperated — wizards often control only selected behaviors, frequently by choosing among prepared options, while other parts run autonomously
• WoZ findings show what an autonomous robot will achieve — they show how people respond to an idealized system, which a real implementation may not match
• Using a wizard removes the need for tight experimental control — wizard timing and judgment vary, so protocols, training and logging are needed to keep conditions consistent

Who should know this

• Human–robot interaction researchers
• Social and service robot designers
• UX researchers prototyping robot behavior
• Ethics reviewers assessing studies that involve deception

Related technologies

• Teleoperation Interfaces
• Shared Autonomy
• Behavior Authoring Tools
• Interaction Logging and Video Annotation
• Participant Debriefing Protocols

Related terms

• Teleoperation
• Godspeed Questionnaire
• Novelty Effect
• Perceived Safety
• Turn-Taking
• Research Through Design

Related projects

• SO-101 Open Robot Arm: 6-DOF robot arm for teleoperation and AI robotics research
• XLeRobot: Practical dual-arm mobile home robot for $660 with <4hrs assembly
• DimOS: Python robotics runtime with agents, SLAM and teleop, no ROS required
• Mobile ALOHA: Bimanual mobile manipulation with 4 robot arms on a wheeled base for teleoperation
• GEAR-SONIC: NVIDIA's whole-body controller for the Unitree G1, trained to track any human motion
• OpenArm: 7DOF humanoid arm for physical AI research in contact-rich environments