Robot control

Researchers: Axel Willekens, Alexander Vandesompele, Victor-Louis De Gusseme, Andreas Verleysen, Rembert Daems, Peter De Roovere, Matthijs Biondina, Joni Dambre, Francis wyffels

We investigate how robots can acquire human-like manipulation skills autonomously in a data-efficient manner. We focus on supplying rich training signals by exploiting prior knowledge embedded in human-task solving and system modeling. We envision structuring robotic learning like human learning to outsource the burden of repetitive tasks to robots so that people can focus on the skills from which they get true joy.

For robots to leave industrial, structured environments and enter the territory of small, dynamic companies and households, we require learning methods that enable autonomous learning with minimal human intervention. In our research, we look at how we can leverage prior information to accelerate learning. We extensively focus on utilizing human knowledge as a primary source of prior information. We employ human task demonstrations, modeling physical systems for gradient-based learning, co-optimization of robot body and brain, and multi-modal instrumentation as a way for scaffolded learning. Consequently, we do not concentrate on one technique for robot control but focus on various techniques varying from classic control to differentiable programming to reinforcement learning.

Our research applications are centered around the manipulation of deformable materials like clothing, biocomposites, fungal foams, and plastics. These materials are more heterogeneous, irregular, and varied than common materials used in robotic applications. The deformable and fragile nature of these materials makes them challenging and interesting for robotics research. We emphasize making our methods deployable in the real world by using virtual, simulated tasks as a tool, not as an end-goal.


  1. Modular Piezoresistive Smart Textile for State Estimation of Cloths
    Proesmans, Remko, Verleysen, Andreas, Vleugels, Robbe, Veske, Paula, De Gusseme, Victor-Louis, and wyffels, Francis
    Sensors 2022
  2. Simpler learning of robotic manipulation of clothing by utilizing DIY smart textile technology
    Verleysen, Andreas, Holvoet, Thomas, Proesmans, Remko, Den Haese, Cedric, and wyffels, Francis
  3. A differentiable physics engine for deep learning in robotics
    Degrave, Jonas, Hermans, Michiel, Dambre, Joni, and wyffels, Francis
  4. Body randomization reduces the sim-to-real gap for compliant quadruped locomotion
    Vandesompele, Alexander, Urbain, Gabriel, Mahmud, Hossain, wyffels, Francis, and Dambre, Joni
  5. Populations of spiking neurons for reservoir computing : closed loop control of a compliant quadruped
    Vandesompele, Alexander, Urbain, Gabriel, wyffels, Francis, and Dambre, Joni
  6. Morphological properties of mass-spring networks for optimal locomotion learning
    Urbain, Gabriel, Degrave, Jonas, Carette, Benonie, Dambre, Joni, and wyffels, Francis