Learning force-based multitrajectory tasks

Leonel Rozo, Pablo Jiménez, Carme Torras

We have developed a complete force-based learning by demonstration (LbD) framework for teaching multi-trajectory tasks to robots. The strengths of this work are many-fold: first, we deal with the problem of learning through force perceptions exclusively. Second, we propose to exploit haptic feedback as a means for improving teacher demonstrations and as a human-robot interaction tool in LbD, establishing a bi-directional communication channel between the teacher and the robot, in contrast to works using kinesthetic teaching. Third, we address the well-known what to imitate? problem from a different point of view, based on the mutual information between perceptions and actions. Lastly, the teachers demonstrations are statistically encoded using a Hidden Markov Model (HMM), and the robot execution phase is developed by implementing a modified version of Gaussian Mixture Regression (GMRa) that uses implicit temporal information from the HMM, which is needed when tackling tasks with multi-valued underlying functions, like ours. Experimental results show that the trained robot is able to accomplish this kind of tasks with a performance comparable to the teachers one and much quicker than trying to reach the goal by chance through random movements.

Overview of the learning system

In the experiments conducted within this framework, the robot learned to extract a ball from a rectangular container by tilting it in different directions, using exclusively force perceptions. The video illustrates the performance of the different parts of the learning system along the experiments.

Video illustrating the learning and execution phase