News Release

Robots learn to follow the crowd – could become digital guide dogs

Reports and Proceedings

Örebro University

Yufei Zhu

image: 

Yufei Zhu.

view more 

Credit: Örebro University

Navigating a bustling railway station or airport is a major challenge for people with visual impairments. New research from Örebro University teaches robots to follow the “unwritten rules” of human movement and make their way through dense crowds. In the long term, the technology could serve as a digital complement to guide dogs.

Today’s navigation robots treat every person as an obstacle to be avoided. When crowds become too dense, they can suffer from what researchers call the “freezing robot problem” and come to a complete standstill because no clear path is available.

Researchers at Örebro University have now developed a new navigation system that enables robots to move through busy public environments. The method is based on a different approach to navigation: People do not stop in a crowd. Instead, we adapt, keep to one side and move with the flow.

“If a robot is to guide a person with a visual impairment through a crowded terminal, it cannot come to a complete stop every time someone walks past. It has to move naturally and follow the flow,” says Yufei Zhu, who recently defended his doctoral thesis in computer science.

Learning location-specific “habits”

In his research, Yufei Zhu developed what are known as Maps of Dynamics. Rather than mapping only fixed features such as walls and doorways, the robot learns how people typically move through a particular location at different times of day.

“You can think of it as a newcomer arriving in a town, observing how local people behave and then copying them. By studying movement flows, the robot learns the patterns and can predict where people are heading,” explains Yufei Zhu.

Tested at Expo 2025 in Osaka

The technology, known as HiCrowd, means that the robot no longer sees a crowd as a threat or an obstacle. Instead, it uses people’s movement patterns as guidance.

The research has been tested in real-world settings at the Science Museum Miraikan in Tokyo and at Expo 2025 in Osaka. The robot, which resembles a small suitcase, was tasked with making its way through dense streams of visitors.

“On one occasion, the robot crossed a crowded hall behind a group of people. When the group turned right, the robot followed them, even though it could not yet see why. A few seconds later, the reason became clear: a large crowd was walking directly towards them. By following the people ahead, the robot safely passed the oncoming flow without having to stop even once,” says Yufei Zhu.

Tests also showed that people nearby neither appeared alarmed by the robot nor kept a distance to it. The robot blended so naturally into the flow of movement that people around it treated it like any other pedestrian.

Could assist travellers at railway stations and airports

One potential future application for the technology is as an assistive device for people with visual impairments. A robot guiding someone through a busy railway station or airport terminal must be able to navigate smoothly without abrupt movements or constantly stopping.

“At a railway station, there are clear habits and movement patterns. Using our methods, a guide robot could lead a person with a visual impairment so that they move naturally with the flow rather than struggling against it,” says Yufei Zhu.

A global research collaboration

The research was carried out at Örebro University under the supervision of Professor Martin Magnusson, in close collaboration with researchers at the Miraikan Accessibility Lab in Tokyo and Bosch in Germany. Through this international partnership, the dynamic mapping technology has been applied in a variety of settings, ranging from warehouse robots in industrial environments to service robots moving among large groups of museum visitors.

Facts about the thesis

  • Title: Spatio-Temporal Representations of Human Dynamics for Motion Prediction and Robot Navigation
  • The research was conducted by Yufei Zhu, a recent PhD graduate in computer science at Örebro University, under the supervision of Professor Martin Magnusson. The project is an international collaboration with the Miraikan Accessibility Lab in Japan and Bosch in Germany.
  • By combining advanced computer science with the Maps of Dynamics and HiCrowd software systems, the technology learns location-specific movement patterns in real time. The method has been successfully tested in real-world environments at the Science Museum Miraikan in Tokyo and during Expo 2025 in Osaka, demonstrating how autonomous robots can guide people and interact naturally with large crowds without interruption.

Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.