Research seminar on: "Learning Conditional Tasks by Demonstration of Multiple Solutions with an Application to Fault Recovery"
Research seminar on: "Learning Conditional Tasks by Demonstration of Multiple Solutions with an Application to Fault Recovery" by Thomas Eiband, Human Centered Assistive Robotics, TUM
Abstract: Certain task structures require to account for varying conditions instead of a pure sequential execution. For instance, spatial sorting of objects requires different goal positions based on the object’s properties, such as weight or geometry. In such cases, human demonstrations can be used to learn the distinctive features of these strategies, which we term solutions. By selecting an initial strategy, termed as nominal solution, deviations thereof can be classified as anomaly, which are detected by monitoring robot pose, external wrench as well as grasp information. Additionally, alternative solutions can be triggered at a state, where the robot shall account for an error autonomously. The method targets on intuitive teaching of conditional tasks and enables the user to transfer recovery behaviors for likely situations.
29 May 2019, 16:30 h Place: ICS Karlstraße 45, 6. Floor, room 6009
For more info please visit: http://web.ics.ei.tum.de/~ehrlich/ICS_seminar/index.htm