Style Transfer for Robotic Arms

This project is part of ongoing research on expressive movement in robotics, at the intersection of engineering, the arts, and rehabilitation. The overall goal is to explore how robotic arms can learn and integrate choreographed and emergent movement qualities, in order to foster richer and more nuanced human-robot interactions. More specifically, the robotic movements will be used within a dynamic play-based rehabilitation framework for patients undergoing therapy.

Building on recent work in expressivity transfer between humans and robots using generative models, this project aims to develop a methodology that allows a robotic arm to adopt expressive motion styles based on human demonstrations. The approach will combine movement analysis tools from dance, biomechanics, and machine learning techniques (variational autoencoders, GANs, attention-based architectures).

Design and implement a style learning methodology that will:

• Extract expressive qualities from human choreographic movements,
• Transfer these expressive qualities to the motion of a robotic arm,
• Experiment with various learning approaches to ensure stability and variability of the generated styles,
• Evaluate resulting engagement both in artistic and clinical rehabilitation contexts, through analytical tools and collaborations with field experts.

Research Tasks

• Collect and annotate motion data from robotic (teleoperation) and human sources (video, depth, and LiDAR).
• Develop machine learning pipelines for style extraction and transfer.
• Conduct experiments on a robotic arm (real-time control, morphological transfer).
• Collaborate with experts in expressive movement analysis to interpret and qualify results.
• Contribute to scientific writing and dissemination in both academic and artistic contexts.