One-shot robot programming by demonstration using an online oriented particles simulation

To provide robots for a wide range of users, there needs to be an easy and intuitive way to program them. This issue is addressed by the robot programming by demonstration paradigm, where the user demonstrates the task to the robot. While there exist a lot approaches that use multiple demonstrations for the learning procedure, single-shot robot programming by demonstration is still a niche. Also, all available approaches in this niche have severe drawbacks. The main contribution of this work is a novel one-shot programming by demonstration approach, that performs an online adaption of a provided trajectory to a new situation. For that, the system regards every sample of the trajectory and every reference (object) in the scene as a particle. These particles interfere with each other by forces and torques that arise from inherent potential fields. Thus, in a new situation the trajectory will adapt to the potential fields of the relocated references and converge to a minimum energy state. We evaluated the approach qualitatively and quantitatively using cross validation.