Statistical, geometrical and dynamic representations of movement

Course description
This course will present various ways of representing movement data and gestures in a mathematical manner, with the goal of analyzing, compressing or generating movements. Several examples of applications will be covered, from generation of manipulation skills in robotics to the analysis of motion capture data. The principle of movement primitives will be presented, which allows reorganization in parallel and in series of ‘motion bricks’, in order to create new gestures or to adapt a gesture to a new situation or to a new kinematic chain. Several movement representations will be covered in the course, arising from different research domains, including statistical modeling (hidden Markov models), differential geometry (Riemannian manifolds) and dynamic systems (dynamic movement primitives).

Objectives

  • acquire an overview of existing techniques, at theoretical, practical and implementation levels
  • cover various examples of applications; examples of implementation in Matlab, C++ and Python will be presented, with source codes provided to the course attendees, which will be exploited to test and explore the techniques described in the course

Academic Instructor
Sylvain Calinon |AIMove’s Academic Director for IDIAP