Dongdong Liu, Yuhang Lin, Alexander Koldy, Vikram Kapila

|Abstract|Code|Paper|Concept|Results|Experiment|Acknowledgment|

Abstract

In recent years, there has been a growing interest in the use of optimal control techniques for diverse application scenarios being envisioned for humanoid robots. Throwing is a common human activity used in a variety of situations, including sports, hunting, rescue, etc. Optimization approaches have previously been considered to maximize the simulated throw performance for humans. Experimental approaches for improving human throw performance have been restricted to observation and mimicry of high performing athletes. In the case of robotics, prior research on throw performance improvement using optimization has been limited to simulations studies, with scant research considering experimental validation. To address this research gap, in this effort, we consider the optimal design, experimental implementation, and analysis of a whole-body throwing motion in a double support humanoid robot with the use of a differential dynamic programming (DDP) approach. We provide results from repeated experiments to demonstrate that the proposed DDP method significantly improves the throwing performance for a 23 degrees-of-freedom humanoid robot in contrast to a previously proposed key-frame (KF) method. With the framework of this paper, a humanoid robot can perform highly dynamic whole-body throwing tasks, achieving optimal throwing distance while maintaining balance stability.

Paper

Concept

The concept of a humanoid throwing motion: (a) initial pose, (b) preparation pose, (c) throw pose, and (d) recovery pose. A ball is released from the robot’s right gripper at the end of the throw phase.

Results

Simulated joint positions for differential-dynamic programming (DDP) approach.
Lower extremity:

Upper extremity:

Framework:

Throw distance comparision for both differential-dynamic programming (DDP) and keyframe (KF) approach

Experiment

(a) A humanoid with encoder-equipped motors mounted on each joint and an inertial measurement unit (IMU) mounted on its pelvis, and (b) the throwing testbed

Timeline snapshots: Throwing experiment with the DDP approach

Timeline snapshots: Throwing experiment with the KF approach

Acknowledgment

Work supported in part by the National Science Foundation under an ITEST grant DRL-1614085, RET Site grant EEC-1542286, and DRK12 grant DRL-1417769. D. Liu thanks his lab colleagues, particularly P. Chauhan, for helping edit early drafts of the manuscript.