Principal Investigator: Mariella Dimiccoli
Email: mdimiccoli@iri.upc.edu
Web: www.iri.upc.edu
Brief Theme Description:
In industrial settings where humans and robots need to collaborate on structured tasks, anticipating human behaviour significantly improves collaboration outcomes. Similarly, in augmented reality (AR) applications, anticipating the future from real-world observations is crucial for smart assistants to help us in an embodied setting or for intelligent systems to perform a task given natural language instructions.
The goal of this project is to develop methods to predict sequences of human actions and human motion based on prior video observations. The focus will be on fully understanding goal-oriented free-style procedures in presence of rich sequence variations, while minimizing labelling efforts. Special emphasis will be given to the role of Language to understand the unfolding of events at both task and motion level.
The outcome of this project will allow to facilitate optimal robot control through informed action planning and to adapt in real-time impedance control for smoother and safer collaboration.
Available Infrastructures: Fully equipped Robot Perception and Manipulation Laboratory, see https://www.iri.upc.edu/research/perception#facilities.
Possible Secondments: University of Freiburg, Germany (Computer Vision and/or Robotics groups). University of Stanford, USA (Robotics group)
Keywords: Human Behaviour Anticipation; Human-Robot Collaboration; Deep Learning.
