Examenspresentation av Hampus Ahlebrand

Examiner: Emmanuel Dean

Opponent: Darshan Gad

Abstract:
Robotics has the potential to solve many problems, such as relieving humans from repetitive and or hazardous work. Thus, robotics receives much research attention. Ideally, humans and robots should be able to seamlessly coexist and collaborate in different environments such as the workplace. However, programming robots with decision-making and control algorithms to handle unstructured environments is difficult. One approach to this problem, when repetitive tasks are involved, is to design intuitive programming interfaces to allow fast reprogramming and thereby utilize human cognitive abilities to effectively increase the robot's adaptability.
The purpose of this thesis is to provide a general programming interface in the Learning from Demonstration paradigm. The main contribution of this work is the formulation of a general approach for allocating hierarchical tasks based on specific task requirements. These tasks can be defined in various spaces, including force, joint and Cartesian position, velocity, and others. The developed interface allows an operator to kinesthetically teach a task that is encoded and generalized using a set of Dynamical Movement Primitives (DMPs). To execute tasks, the operator provides a hierarchical task specification from which a controller is automatically synthesized and DMPs prioritized. The method is validated in both simulation and experiment.

Welcome!

Hampus and Emmanuel