Work Packages

Objectives:

• Development of representations of everyday activities in general as well as specializations for observations of humans performing tasks and logged robot task executions
• Development of formal representation of knowledge and data needed for robotic cooking and kitchen tool manipulation skills
• Methods for storing these formal robot skill representations, for reasoning upon them, for combining, learning, and constraining them, and for transforming them into executable robot programs

Leading partner institution: Uni Bremen

Objectives:

• Extraction and symbolic representation of the following information from a demonstration video:
  • 3D geometric and appearance properties of scene objects
  • 3D configuration of human actors manipulating objects
  • association of involved entities over time

Leading partner institution: FORTH

Objectives:

• Development of an action and movement interpretation system that generates fast, smooth and dynamically adequate movements for constraint- and optimization-based action specifications
• Mapping of “low level”, generic infrastructure into a formal, more application-directed representations (so called domain-specific languages) that can be reasoned with, that afford transformation into plans, and that can be used integration
• Develop schedulers that execute the right pieces of code, at the right moment, and at the right level of abstraction

Leading partner institution: KU Leuven

Objectives:

• Extract the information about the objects in the environment (geometric, appearance, kinematics and dynamic properties) using different sensors (vision, force, distance, tactile)
• Provide input to control loops for the execution of tasks based on multi-sensory feedback according to therobotic platform capabilities (single, dual-arm, whole-body) for known objects
• Provide input for learning of task constraints and task adaptation based on the success of the execution, thus enabling robust performance over time

Leading partner institution: KTH

Objectives:

• Learning of task constraints as a bootstrapping step prior to learning more complex motor tasks
• Learning adaptive stiffness control; this includes learning the dynamics of arm, hand and finger motion that ensure grasp stability, to learning how to adapt this motion when in interaction with the object
• Learning of haptic interaction; this task extends the previous task by learning how to adapt the dynamics of arm motion when in physical interaction with another human via an object

Leading partner institution: EPFL

Objectives:

• Development of the following components:
  • A plan language to represent and specify the desired behaviour of the robot
  • Prediction methods to infer gaps in incomplete symbol action specifications
  • Transformation methods to automatically modify plans to forestall execution errors
  • Mechanisms for flexible monitoring of task execution

Leading partner institution: Uni Bremen

Objectives:

• Illustration of the successful integration of the research and development in the project with real benchmarking
• Assessment of potential operation in real conditions using the various robotic platforms available within the project
• Building a distributed multi-level architecture that integrates knowledge and development resulting from other workpackages with a strong link to representation, learning, and robotic representation and control

Leading partner institution: CNRS

Objectives:

• Foster participation in the scientific community, among industry stakeholders, and potential user groups
• Outlining of future industry-oriented dissemination activities
• Assessment of the results achieved for patenting and prototyping and future commercialization

Leading partner institution: Aldebaran

Objectives:

• Ensures efficient and effective coordination and management of the consortium
• Organization and management of multi-partner projects
• Ensure quality management and quality control of deliverables and reports

Leading partner institution: Uni Bremen