Effective Leaning of Social Affordances for human-robot interaction – ELSA

Affordances are action opportunities directly perceived by an agent to interact with its environment. The concept is gaining interest in robotics, focusing on the potential interactions of the robot with objects rather than the sole physical properties of the environment. In this project, we aim to extend this notion to social affordances which have not been really investigated in robotics. Here we make the hypotheses (1) that robots can learn social affordances in the same way as they learn action affordances in non-social situations; (2) that robots that can autonomously recognize social affordances offered by humans will more efficiently and appropriately respond to the human, thus facilitating human-robot coordination and cooperation.

Thus, we will investigate the following problems related to social affordances in robotics:
How can robots better take into account the way humans perceive and use social affordances of other agents? Which information signals the availability of an affordance? Which behavior enables one to use social affordances to produce efficient interactions? (WP1) How to integrate the social affordances within a modular decision-making and reinforcement learning architecture for robots? How should a world model be designed, so as to encompass the variability of the human reactions to robot solicitations? (WP2) How should a robot behave to efficiently use the social affordances offered by other agents? Which information is important to represent the effects of the interaction accurately so that the robot can incrementally increase its action repertoire in social contexts? (WP3) How to consider and take advantage of social affordances (incl. their learning) in an architecture for HRI? What needs to be added/changed to decision-making and task planning? (WP4)

Our project addresses a critical lack of basic knowledge regarding how social affordances are learned and used in the context of Human-Robot Interaction. Its major strengths are: i) Addressing an relatively unexplored subfield (social affordances in robotics) that rely on well-studied fields (affordances in robotics but also in psychology); ii) Being in the continuity of earlier works from the consortium, including joint publications (projects RoboErgoSum and IMAGINE); iii) Being highly interdisciplinary and involving complementary expertises. Altogether, the scientific impact of our project will be in terms of publications in high impact-factor journals, high standard international conferences, scientific reviews, seminars and workshops organised all along the project.
The consortium is composed of two research institutions in France and two in Austria. The French partners include: Mehdi Khamassi (ISIR, project coordinator), Benoît Girard (ISIR, WP2 leader), Aurélie Clodic (LAAS, WP4 leader). The Austrian partners, Justus Piater (IFI, Austrian project coordinator, WP3 leader), Erwan Renaudo (IFI, WP3 post-doc researcher) and Mathias Schurz (DiSC, WP1 leader).