Movement templates for learning of hitting and batting

Abstract

Abstract-Hitting and batting tasks, such as tennis forehands, ping-pong strokes, or baseball batting, depend on predictions where the ball can be intercepted and how it can properly be returned to the opponent. These predictions get more accurate over time, hence the behaviors need to be continuously modified. As a result, movement templates with a learned global shape need to be adapted during the execution so that the racket reaches a target position and velocity that will return the ball over to the other side of the net or court. It requires altering learned movements to hit a varying target with the necessary velocity at a specific instant in time. Such a task cannot be incorporated straightforwardly in most movement representations suitable for learning. For example, the standard formulation of the dynamical system based motor primitives (introduced by Ijspeert et al.[1]) does not satisfy this property despite their flexibility which has allowed learning tasks ranging from locomotion to kendama. In order to fulfill this requirement, we reformulate the Ijspeert framework to incorporate the possibility of specifying a desired hitting point and a desired hitting velocity while maintaining all advantages of the original formulation. We show that the proposed movement template formulation works well in two scenarios, i.e., for hitting a ball on a string with a table tennis racket at a specified velocity and for returning balls launched by a ball gun successfully over the net using forehand movements. All experiments were carried out on a Barrett WAM using a four camera vision system.

Estimating the non-linear dynamics of free-flying objects

“…A body of work has been devoted to autonomous control of fast movements such as catching [3][4][5][6][7][8][9][10], dynamic re-grasping (throwing an object up and catching it) [11], hitting flying objects [12,13] and juggling [14][15][16]. Most approaches assumed a known model of the dynamics of motion and considered solely modeling the translational object motion.…”

Learning throwing and catching skills

“…Furthermore, DMPs guarantee the stability of the movement generation. In [7], we have introduced a modification of the DMPs that allows to generalize them to arbitrary velocities and positions at the end of the movement. Moreover, DMPs are linear in parameters and, hence, it is straightforward to learn a movement by imitation learning [6] and to refine it by reinforcement learning [8].…”

A probabilistic approach to robot trajectory generation

“…Movement primitives (MP) [1], [2], [3] are considered a state-of-the-art-approach for learning robot movement generation. They have been used successfully to solve many complex tasks, including the 'Ball-in-the-Cup' game [4], Ball-Throwing [5], Pancake-Flipping [6] and bipedal gait generation [7].…”

“…The DMP framework is well-established in robotics as they are straightforward to obtain from imitation [1] and reinforcement learning [10], can be used for both rhythmic and stroke-based movements, the movement can be generalized to different final-positions, and the movement speed can be adjusted by a temporal scaling parameter. The original DMP framework has been extended in [3] to generalize also to different final velocities and in [6] to use a timevarying spring-damper system which effectively modulates the stiffness of the executed movement.…”

“…Typically, the generalization to new final positions or velocities is based on heuristics and cannot be directly obtained by imitation learning. While some of the limitations of DMPs have been fixed by some extensions [8], [3], [5], such extensions are typically based on heuristics or only fix single limitations of the DMPs. For example, multiple MPs can be combined [8], but it is unclear how this combination affects the resulting motion.…”