Robot Speed vs. Accuracy: How Users Perceive Intention Recognition

What is it about?

When robots move around people, they use AI models to infer our intentions—like whether we are walking toward them to talk or just passing by. Some of these models are very fast but prone to mistakes, while others are slower because they take more time to "think" to achieve better accuracy. This study examines how these technical trade-offs affect how people actually feel during the interaction. We found that being fast isn't enough; if a robot reacts quickly but guesses wrong, people find it unresponsive and confusing. Surprisingly, users didn't always prefer the most accurate model if it was too slow to respond. The best experience came from a balance where the robot was reliable enough to be predictable, which made people feel safer and more comfortable. Essentially, for a robot to feel like a "social" partner, it needs to be "smart" enough to understand us before it tries to be fast.

Featured Image

Photo by Katja Ano on Unsplash

Photo by Katja Ano on Unsplash

Why is it important?

While technical benchmarks often measure how well AI models perform in isolation, our work is unique because it shifts the focus from "how accurate is the model?" to "how does the model make the human feel?". As robots increasingly enter human-shared spaces, understanding this psychological impact is timely and essential for creating machines that people actually trust and want to interact with. A significant finding of our research is that a robot's speed and accuracy improve the user experience only if they lead to behavior that is easy for a human to predict. We discovered many insights. Objective speed alone does not make a robot feel more responsive; it requires a baseline of accuracy to be perceived as effective. Perceived fluency in an interaction is driven more by the robot's understanding of human intent than by its reaction time. Predictability is the most critical factor, acting as the primary bridge that translates technical performance into feelings of comfort and safety for the user. By identifying that moderate latency is acceptable if it yields higher reliability, we provide a practical roadmap for developers to prioritize "social intelligence" over raw speed. This helps ensure that the next generation of mobile robots is not just faster, but more harmoniously integrated into our daily lives.

Perspectives