A team of experts from Imperial College London (UK) has created a new system for robots that uses game theory , allowing the robot to learn how and when to help a human being.
The last decade has seen robots increasingly work with humans, whether in factories, in assistive devices for people with physical disabilities, or in the field of surgery. However, robots currently cannot react in a personalized way to individual users, limiting their usefulness to humans.
Now, Professor Etienne Burdet and his team have developed the first interactive robot system that captures the behavior of the human user and uses it to predict their next movements.
Reactive system
The experts created a reactive robotic programming system that allows a robot to continuously learn the movements of the human user and adapt its own movements accordingly.
“By looking at how humans physically interact with each other, we found that they are very successful through their unique ability to understand the control of others. We apply a similar principle to design human-robot interaction ,” Burdet explains.
The research, conducted in collaboration with the University of Sussex and Nanyang Technological University in Singapore, has been published in the journal Nature Machine Intelligence.
Normally, humans can only control robots using fairly simple techniques , either in ‘master-slave’ mode, where the robot reproduces or amplifies the human’s movement, as with an exoskeleton, or where the robot does not consider the human user in absolute, as in rehab. Humans are unpredictable and constantly change their movements, making it difficult for robots to predict human behavior and react in useful ways. This can lead to errors when completing tasks.
The researchers set out to examine how a contact robot should be controlled to provide a stable and adequate response to a user with unfamiliar behaviors to activities such as sports training, physical rehabilitation or driving.
Games theory
The authors developed a robot controller based on game theory. Game theory consists of multiple players competing or collaborating to complete a task. Each player has their own strategy and all players try to optimize their performance, while assuming that their opponents will also play optimally. To successfully apply game theory to their interaction, the researchers had to overcome the problem that the robot cannot predict human intentions by reason alone.
They used game theory to determine how the robot responds when interacting with a human, using the difference between its actual and expected movements to estimate the human’s strategy: how the human uses errors to generate new actions. For example, if the human’s strategy does not allow them to complete the task, the robot can increase its effort to help. Teaching the robot to predict the strategy of the human being allows it to change its own response. If the robot can estimate the human’s strategy, it can change its own strategy in response.
The team tested their new system in simulations with human subjects, showing that the robot can adapt when the human’s strategy changes slowly, as if the human is regaining strength, and when the human’s ability is changing and inconsistent, such as after sustaining an injury. .
Working in harmony
“By allowing the robot to identify human user behavior and exploiting game theory to allow the robot to react optimally to any human user, we have developed a system where robots can work in much better harmony with humans.” , concludes Yanan Li, co-author of the study.
Reference: Differential game theory for versatile physical human – robot interaction. Y. Li, G. Carboni et al. Nature Machine Intelligence, volume 1, pages36–43 (2019). DOI: https: //doi.org/10.1038/s42256-018-0010-3
