Europe and Japan both face problems of shrinking and aging population. This raises several issues, like missing workforce, and robots are often seen as a possible way for alleviating them. However, that would require that the robots are introduced into human societies and work in direct contact with people.?
The field of Human-Robot Interaction (HRI) is concerned with the ways in which robots should interact with people in the social world. It aims to realize robots that will be present nearby us, humans, and collaborate with us, support us or entertain us. But dealing with humans is notoriously difficult, and bringing this vision into reality proved to be a huge challenge. At the same time, nowadays there are many successful applications of Artificial Intelligence (AI) techniques to increasingly complex tasks. Regardless of that, the field of HRI has largely not yet taken benefit from advanced AI techniques. Conversely, HRI can be considered as a very motivating challenge for AI, where the human is physically and mentally in the loop.
The goal of the AI4HRI project will be to develop and tightly integrate several crucial AI technologies that will allow social robots to appropriately and skillfully deal with humans around them. The first skill that we believe is currently missing in HRI are knowledge management methods and models, as well as knowledge reasoning abilities that can be used by a robot interacting with humans. These abilities have already been developed for situations where robots act alone. In the scope of AI4HRI we will ascertain the necessary scientific advances that need to be done to be able to take into account interaction with humans. We will answer these fundamental questions: what needs to be implemented successfully in the knowledge management system, which kind of models of the humans the robot should build, and which kind of reasoning abilities the robot should have regarding itself and the human it interacts with.
Building upon this knowledge, a second skill that we will study is how a robot needs to socially interact with a human. Explicit manual programming of the interaction behaviors that the robot should execute is what is often done in HRI, but it is difficult to create interactions that are robust to all variations of environment or human behavior. Therefore we will use machine learning techniques to learn the interaction patterns that humans use in interactions with other humans and copy them to a human-robot interaction context.
A third skill that is essential for socially interacting robots but still not well developed at this point is to be able to cooperatively work with a human on the same task. We will thus work on appropriate techniques for the planning and execution of joint actions by taking advantage of the knowledge management and reasoning abilities described above.
Most importantly, these AI abilities for HRI will be tightly combined into a single architecture, an autonomous decision-making system that allows high-level interactions of a robot with human users. The whole architecture will be open-source and we plan to share it with other researchers. Moreover, the architecture will be validated in both interactions through virtual reality, as well as in a real-world implementation on a social robot in a public space.
The AI4HRI project brings together three teams who are not only world leaders in AI or HRI, but who also have very complementary approaches to solving the issues addressed by the project. University of Bremen (Germany) has extensive expertise on knowledge management and reasoning, Kyoto University (Japan) has large know-how in learning for social interactions, and LAAS-CNRS (France) will add to that strong expertise in human-aware design and in joint action planning and execution.
We are confident that the result of this project will become an important milestone for achieving a future with socially interactive robots working in close vicinity with us.