Learning to Serve: An Experimental Study for a New Learning From Demonstrations Framework

Abstract

Learning from demonstrations is an easy and intuitive way to show examples of successful behavior to a robot. However, the fact that humans optimize or take advantage of their body and not of the robot, usually called the embodiment problem in robotics, often prevents industrial robots from executing the task in a straightforward way. The shown movements often do not or cannot utilize the degrees of freedom of the robot efficiently, and moreover can suffer from excessive execution errors. In this letter, we explore a variety of solutions that address these shortcomings. In particular, we learn sparse movement primitive parameters from several demonstrations of a successful table tennis serve. The number of parameters learned using our procedure is independent of the degrees of freedom of the robot. Moreover, they can be ranked according to their importance in the regression task. Learning few parameters, which are ranked, is a desirable feature to combat the curse of dimensionality in reinforcement learning. Real robot experiments on the Barrett WAM for a table tennis serve using the learned movement primitives show that the representation can capture successfully the style of the movement with few parameters.

Towards Orientation Learning and Adaptation in Cartesian Space

“…It is well known that robot trajectories should be smooth in order to facilitate the design of controllers as well as the execution of motor commands [16], [17]. For instance, in a table tennis scenario that needs fast striking motions, extremely high accelerations or jerks may degrade the final striking performance, given the physical limits of motors.…”

“…For instance, in a table tennis scenario that needs fast striking motions, extremely high accelerations or jerks may degrade the final striking performance, given the physical limits of motors. It is possible to formulate this constraint as an optimization problem and search for the optimal trajectory via an iterative scheme, as done in [16]. In this section, we consider the problem of learning and adapting quaternion trajectories while taking into account jerk constraints.…”

“…It is worthy pointing out that many imitation learning ap- proaches focus on mimicking human demonstrations, whereas constrained skill learning is often overlooked. As discussed in [16], [17], trajectory smoothness (e.g., acceleration and jerk) will influence robot performance, particularly in time-contact systems (e.g., striking movement in robot table tennis). Thus, it is desirable to incorporate smoothness constraints into the process of learning orientations.…”

An improved approach of task-parameterized learning from demonstrations for cobots in dynamic manufacturing

“…tennis playing [6] and industrial assembly [7]. For the above pick-and-place task, using TP-LfD, some human's demonstrations are taken to train a cobot to act under different task settings.…”