TFG

Variable Impedance Control with Movement Primitives in Latent Spaces

Work default illustration

Information

  • Started: 01/02/2019

Description

The aim of this project is to understand, test and improve the stochastic representation of robot motion, which provides us with a time-varying precision requirement. This precision can then be used as a reference for setting the stiffness/rigidity with which a robot moves - a crucial element in human-robot interaction -. Then, for learned tasks, we can separate the task space into a relevant part (which needs to be controlled for task fulfilment) and a secondary part (which can be exploited for secondary goals). We will work on both of them in order to obtain a better controller for human-robot interaction in tasks like placing a scarf around a mannequin -or a person's- neck. See http://www.iri.upc.edu/groups/perception/#ScarfTask

Profile of the student:
We are looking for a Grau student from mathematics/mechanical/industrial engineering, or a Master student; with an interest in control, machine learning, and robotics. This work is intended as a TFG/TFM and will aim to publish an article at an international robotics conference.

Some valuable skills are:
- Linear Algebra: Matrix and vector operations, eigenvalue/vectors manipulation and understanding.
- Probability and Statistics: Statistical inference with Gaussian distributions. Understanding of Bayesian inference and Expectation Maximization methods is a plus.
- Mechanics and Control: Understanding of a mechanism's equations of motion, basic control schemes, etc.
- Programming in C++ and/or Matlab

Remuneration
This work can be remunerated through an INIREC grant. This grant is expected to have a duration of 6 months, with a dedication of 20 hours per week, and a remuneration of around 500 € per month.

How to apply
Send an e-mail to jobs@iri.upc.edu with your CV and academic transcripts.

The work is under the scope of the following projects:

  • LatCom: Variable Impedance Control with Movement Primitives in Latent Spaces During Compliant Manipulation (web)