Robotics Research - Sensor-based Motion for Autonomous Robots

This research is one of the robotics research activities of the Department of Mechanical Engineering at the Katholieke Universiteit Leuven in Belgium. The group's research activity is concerned with the “intelligent” use of sensor information in tasks executed by one or more robot arms.

The group is headed by professors Joris De Schutter and Herman Bruyninckx, supervising about half a dozen PhD students.

In our research, robot motion is in compliance with the constraints imposed by the data from the sensors: compliant motion. Since about three decades, our main expertise lies in force-controlled compliant motion, but the global information provided by distance and vision sensors is complementary to the local information provided by a force sensor. This research targets both industrial robots and service robots.

• Using “intelligent” sensor processing to improve the autonomous task execution capabilities of robot arms. The robot motion is in compliance with the constraints imposed by the data from the sensors: compliant motion. Our main expertise lies in force-controlled compliant motion, but also in using distance and vision sensors. Our approach is mainly based on Bayesian probability theory: Kalman filters, particle filters, Bayesian networks, Markov random fields, etc.

• To make increased complexity of robot arm tasks feasable, by developing a uniform task specification framework, together with the necessary software support to provide an operator-friendly interface. Constraint-based programming is the key to this flexibility: constraints scale and compose much better than the traditional setpoint-based robot specification approaches.

We work towards a better understanding of, and more software support for, what we call Autonomous Compliant Motion, that is, robots that extensively use “intelligent” sensor processing to achieve tasks in “contact” with their environments. “Intelligent” sensor processing means: using Bayesian information processing and estimation techniques, such as Kalman filters, particle filters (“sequential Monte Carlo”), or rather simple, realtime Bayesian networks for localization or recognition. “Contact” means: physical contact in which forces are generated, controlling the distance to the environment, or the relative position and orientation between robot and (moving) objects.

We are integrating all necessary know-how into a single, large-scale support framework, to offer a unified approach towards task specification, sensor processing and (hard) realtime control. The components within this framework that have been developed so far, or that are currently under active development, are:

• Research
• Publications
• People
• Software
• Jobs

People

PhD students

• Wilm Decré
• Tinne De Laet
• Ruben Smits
• Diederik Verscheure
• Koen Buys
• Erwin Aertbeliën
• Zhengdong Wang

Postdocs

• Peter Slaets

Alumni

• Kasper Claes, now at Materialise, one of our Department's spin-off companies.
• Wim Meeussen, currently with Willow Garage.
• Johan Rutgeerts, currently with Van De Wiele
• Peter Soetens, currently with FMTC
• Klaas Gadeyne, currently with FMTC
• Walter Verdonck, currently with FMTC
• Tine Lefebvre, currently with Atlas Copco
• Ernesto Staffetti, Post-doc 2000-2001. Researcher at Escuela Superior de Ciencias Experimentales y Tecnología, Universidad Rey Juan Carlos, Madrid, Spain.
• Lyudmila Mihaylova, Post-doc 2000-2002. Research Fellow at the Department of Electrical and Electronic Engineering, University of Bristol, UK.
• Mourad Oussalah, Post-doc 1999-2000, Lecturer at the School of Engineering, The University of Birmingham, UK.
• John Lloyd, Post-doc 1998-1999. Currently with University of British Columbia, Canada.
• Stefan Dutré, currently with LMS
• Jan De Geeter, currently with Metris
• Johan Baeten, professor at the Katholieke Hogeschool Limburg, Diepenbeek, Belgium
• Sabine Demey, currently with Materialise
• Wim Witvrouw
• Dirk Torfs
• Bagus Made Arthaya
• Jayantha Katupitiya, Post-doc 1995. Senior Lecturer at the School of Mechanical and Manufacturing Engineering, The University of New South Wales, Sydney, Australia.

Orocos control software

When, around 2000, we were confronted with aging controller hardware and software for our research infrastructure, we decided to start a very ambitious, long-term software development project. The project is based on realtime Linux, and is creating a large framework of software components to build advanced control applications. The framework is released under a Free Software license, allowing free use for academic as well as industrial applications. Orocos is subdivided in three libraries. One handles the realtime processing, one the kinematics, and the last one bayesian filtering. More information can be found on the Orocos web pages.

Jobs, Theses

Please, contact Joris De Schutter or Herman Bruyninckx.