Godspeed Questionnaire

• Also called: Godspeed Questionnaire Series, GQS
• Category: Control › Human in the loop
• Tags: human-robot-interaction, user-studies, evaluation, questionnaires, social-robots

A set of five questionnaire scales from Bartneck et al. (2009) that measure how people perceive a robot: anthropomorphism, animacy, likeability, perceived intelligence and perceived safety.

Definition

The Godspeed questionnaire is a set of standardized rating scales for measuring how people perceive robots, introduced by Christoph Bartneck, Dana Kulić, Elizabeth Croft and Susana Zoghbi in 2009 in the International Journal of Social Robotics. It has five subscales: anthropomorphism, animacy, likeability, perceived intelligence and perceived safety. Each item is a five-point semantic differential, a pair of opposite words such as Fake–Natural or Incompetent–Competent, and participants mark where the robot falls between them. Subscale scores are usually the mean of their items.

The perceived safety subscale has only three items, Anxious–Relaxed, Agitated–Calm and Quiescent–Surprised, and asks participants to rate their own emotional state rather than the robot. Because it is short, easy to administer and has been translated into many languages, Godspeed became one of the most widely used instruments in human-robot interaction research, and its subscales are often used on their own.

Its popularity has also drawn criticism. Some word pairs are not clear opposites, the subscales are correlated with each other, and the original validation was limited by today's standards. The three-item safety subscale in particular captures general arousal and calm rather than the specific sense of danger, which has motivated newer, dedicated scales for perceived danger and safety. Researchers are encouraged to report reliability for their own sample and to choose instruments that match the construct they actually want to measure.

Why it matters

Godspeed scores appear in a large share of HRI papers, so engineers reading or running user studies need to know what its subscales measure and where they fall short. Using it well, or choosing a better-matched instrument, makes study results more comparable and more meaningful.

Key applications

• Comparing robot appearance or behavior conditions in user studies
• Measuring anthropomorphism and likeability of social robots
• Assessing perceived intelligence of robot behaviors
• Quick screening of perceived safety in interaction studies
• Benchmarking results across HRI studies that use the same scales

In practice

An HRI team compares two gaze behaviors on a social robot that greets visitors. After each interaction, participants fill in the likeability, perceived intelligence and perceived safety subscales of the Godspeed questionnaire, presented in the participants' own language. The team computes subscale means, reports internal consistency for their sample, and tests the difference between conditions. Because they are specifically interested in whether a robot that stares makes people uneasy, they also add a dedicated perceived danger scale and record how far visitors stand from the robot, rather than relying on the three Godspeed safety items alone.

Common misconceptions

• Godspeed is a safety questionnaire — perceived safety is only one of its five subscales, and the smallest
• The scales are fully validated for every use — reliability and validity should be checked for each study population and translation
• Godspeed measures how safe a robot is — it measures participants' self-reported feelings, not physical safety

Who should know this

• HRI researchers designing and reviewing user studies
• Engineers evaluating social and service robot behaviors
• Students running their first robot user study

Related technologies

• Semantic Differential Scales
• Likert Scales
• Robotic Social Attributes Scale (RoSAS)
• Negative Attitudes Toward Robots Scale (NARS)
• Cronbach's Alpha

Related terms

• Perceived Safety
• Novelty Effect
• Humanoid Robot
• Research Through Design

Related projects

• evo: De-facto standard tool for evaluating and plotting odometry/SLAM trajectories
• Reachy Mini: Open-source desktop robot from Pollen and Hugging Face with an expressive 6-DoF head
• RoboDojo: Manipulation benchmark that runs in Isaac Sim and on real arms
• Inspect Robots: Define a robotics benchmark once, run any policy on any robot
• SenseNova-RoboRSI: Open runtime that runs, scores and improves robot agents driven by language models
• RAI: Vendor-agnostic agent framework for ROS 2 robots and physical AI