The robot gastronomy project is developing methods for robots to model and learn manipulation skills for preparing food, such as salads. Key challenges of this work include modeling and generalizing over material properties as well as reasoning about deformable objects.
The Pillar project explores synergies between planning and learning to create a fundamental framework for lifelong learning in robotics. Our research focuses on generating novel subtasks and skills, as well as determining how to structure the skill learning process.
Vibrations are an important source of information for performing and monitoring manipulation tasks. We are therefore developing methods for actively generating and interpreting vibration signals. Our tasks include manipulating granular materials such as coffee and soil.
Manipulation tasks can be modeled as hybrid systems with piecewise continuous dynamics corresponding to distinct modes. Mode switches are often the result of making and breaking contacts. Our research focuses on learning the mode structure of manipulation tasks and using this structure as the basis for autonomous skill learning.