Short description of the project:
Deformable objects are encountered in many industrial applications, ranging from foams and padding of variable stiffness to plastic parts that need to be snapped into a fixture, cartons that need to be folded and packed tightly, etc. Similarly, there are numerous potential applications of deformable object manipulations in fields ranging from agriculture to medicine such as fruit picking and suturing.
This project aims to focus on further developing machine learning and control theory foundations of modeling and predicting a deformable object's interactions with a robot directly from available vision and force sensor data. The project's goal is to develop models learned from sensor data which map the robot actions to the expected changes of the environment and the deformation state of the object while respecting physical constraints. Such models are a key ingredient for model-based control and model-based reinforcement learning in the context of robotic manipulation. The project will focus in particular on objects that are deformable but exhibit some degree of stiffness, such as plastics, metal sheets, etc.
-- One position is advertised at Chalmers University of Technology, Gothenburg, Sweden under supervision of Yiannis Karayiannidis, Systems and Control, Electrical Engineering, Chalmers University of Technology. The position at E2/Chalmers will have an emphasis on data driven modeling of robot-object interaction and design of control policies based on reinforcement learning principles.
The position at KTH will have an emphasis on developing novel mathematically rigorous ideas for machine learning methods that can integrate constraints posed by known physics with state of the art machine learning methods (Deep Learning and/or Variational Inference). Applications from candidates with a wide variety of backgrounds besides computer science/robotics and including pure/applied mathematics, statistics and physics are encouraged to apply.