The problem of communication and understanding between human beings may be a big issue in every person's mind. And the same communication barrier exists between robots and humans, and cooperation between humans and machines is an important aspect of automation in an ever-changing world of technology.

Recently, computer scientists at the University of Bristol have developed a new method of human robotics research, similar to human counterfactuals, which they used in a handheld robot that would first make predictions about the next step in human behavior and then frustrate the user by betraying his or her plans, thus enabling this handheld robot to demonstrate and learn an understanding of human intentions.

This intelligent handheld robot can collaborate with the user to accomplish tasks. In contrast to traditional power tools, which know nothing about the task they perform and are completely under the control of the user, the handheld robot has knowledge about the task and can help by guiding, fine-tuning movements and determining the task sequence.

This new research suggests that deliberate frustration is part of the process of developing robots that work better with users. While this can help to complete tasks faster and more accurately, users may become annoyed when the robot's decisions don't match their own plans.

The latest research in this area by Janis Stolzenwald, a PhD candidate in the Department of Computer Science at the University of Bristol, and Professor Walterio Mayol Cuevas, explores the use of intelligent tools that can bias a user's decision-making based on their intentions. The research is a novel and interesting twist on the study of human robots, which aims to first anticipate users' needs and then go against those plans.

Professor Mayol-Cuevas said, "If you are frustrated with a machine that could otherwise help you, it is easier to recognize and measure than the usually elusive signals of human-robot cooperation. If users get frustrated against their plans when we instruct the robot to perform an action, we can give the robot a better idea of what they want to do."

"Just as short-term predictions of each other's behavior are essential for successful human teamwork, our research shows that integrating this capability into collaborative robotic systems is critical for successful human-robot collaboration." In this study, the researchers used a prototype that tracked a user's line of sight and used machine learning to obtain short-term predictions about expected movements, and then used that knowledge as the basis for robot decisions (such as the next move).

The Bristol team trained the robot for the study using more than 900 training examples from a participant's pick-and-place task, and the study centered on the evaluation of an intentional prediction model. The researchers tested the robots in two situations: obedience and rebellion. The robots were programmed to follow or defy the user's intended intentions. Knowing the user's goals gave the robot the ability to rebel against their decisions. The difference in frustration response between the two situations proved the accuracy of the robot's predictions, thus validating the intent prediction model.

Janis Stolzenwald, a PhD student funded by the German Academic Scholarship Foundation and the UK's EPSRC, conducted user experiments and identified new challenges for the future. He says: "We found that intent modeling is more effective when visual data is combined with task knowledge. This raises a new research question: how do robots retrieve this knowledge? We could conceivably learn from a demonstration or involve another person in the task."

To address this new challenge, researchers are currently investigating shared control, interaction, and new applications in studies on telecollaboration via handheld robots. Maintenance tasks act as user experiments in which handheld robot users will be assisted by experts who remotely control the robot.