CoHRoS - Collaborative Programming of Highly Redundant Robots

A CLOOS robot system during welding of a large work pieceNowadays robotic systems are used for industrial applications like welding, grinding or varnishing of large work pieces, e.g. during the production of earthmoving equipment, agricultural machines or automobiles. These systems are highly redundant with up to 10 or more degrees of freedom allowing multiple configurations and a wide range of possible movements. However, programming these robots is tedious, costly and needs highly specialized expertise, which is an important factor to achieve a reasonable return on investment for automation.

The ECHORD++ experiment CoHRoS intends to develop a practical and robust method for assistive teaching, meaning the robot will learn and generalize from few demonstrations provided by the programmer. The results will redefine and advance the state-of-the-art programming for highly redundant (high degree of freedom) robot systems. The experiment will contribute in adapting and advancing methods to structure interactive programming for those systems where is not easily possible that a human is performing the motions which the robot then has to repeat.

Software Integration for Real-time Control of CLOOS robots

The project’s approach is to treat the whole robotic system including external axes as one highly-redundant kinematic chain, where the inverse kinematic's redundancy resolution includes all axes including the positioning systems. On order to solve this redundancy, machine learning approaches need to be applied to learn from human demonstrations. It is a particular expertise of Cloos as a producer of 7-DOF welding robots to integrate them with additional positioner axes into complete robotic welding systems utilizing a proprietary safety-integrated system controller.

However, a major challenge in this project is to integrate all the required different components such as machine learners, interaction controllers, inverse kinematics modules in a real-time capable software architecture together with the CLOOS robot controllers. 

 

Efficient Teaching and Exploitation of Redundancy Resolutions

In contrast to previous projects on physical human-robot interaction, the highly redundant robot systems in this project are not suited for direct physical human-robot interaction. Instead the user interface relies on simple keyboard-based control like with traditional industrial robots. As the project's main goal is to simplify programming of these systems, a key challenge is how to structure the user interaction for providing demonstrations with this interface. Practically, in the first stage the programmer needs to provide a few demonstrations for feasible redundancy resolutions which are learned with machine learning techniques and generalized to the significant workspace. In the next stage, the task teaching for the end-effector or further redundancy resolution teaching will be supported through an assisted mode. A hierarchical controller resolves the redundancy of the external positioning systems and/or redundant axes of the typically 7-DOF robot as secondary objective based on the generalization from the learned configurations. 

Partners

Contact:
Prof. Jochen Steil
Dr.-Ing. Christian Emmerich