Deep-learning in virtual reality for effective human-robot group collaboration

Robotics is now extending to broader applications where robots play social roles at home, in schools or hospitals.

To be successful in these new roles, robots need to be equipped with social skills that facilitate effective collaboration with humans.

This project will develop new computational approaches, based on deep-learning and virtual/mixed reality, to model and evaluate agent multimodal behaviour coordination (e.g., navigation, body orientation, full-body motions, eye-gaze, speech) with small groups of people.

The aim of the model is to enable effective and natural approach, interaction initiation and group collaborations between a robot and multiple humans.

The project will: (1) use a new methodology based on virtual reality and virtual robots to conduct efficient prototyping and learning of behaviours prior to the use of physical robots (milestone 1); (2) develop socially-aware methods for behaviour coordination that effectively enable a robot to navigate to and position itself within a group, recognising when it is appropriate or not to join an ongoing interaction (milestones 2 and 3); and (3) leverage mixed reality to conduct human-robot interaction studies assessing the impact of the methods and embodiments on users’ quality of interaction, acceptance and safety (milestone 4).

The project will open new frontiers and opportunities, making a first step towards the development of artificial systems capable of better collaboration in human-robot teams.