Visual Perception of Humanoid Movement
Frank E. Pollick1'2 Joshua G. Hale2’3 Phil McAleer1
1 University of Glasgow, Department of Psychology
58 Hillhead Street, Glasgow G12 8QB, UK
{frank, phil}Qpsy.gla.ac.uk http://www.psy.gla.ac.uk/~frank
2 ATR Computational Neuroscience Laboratories,
Department of Humanoid Robotics and Computational Neuroscience
2-2-2 Hikaridai, Seika-cho, Soraku-gun, Kyoto 619-0288
3 University of Glasgow, Department of Computer Science
Glasgow G12 8QQ
[email protected] http://www.dcs.gla.ac.uk∕~hale j
Abstract
Wc examined similarity judgements of arm
movements generated by different control
strategies with the goal of producing natu-
ral looking movements on humanoid robots
and virtual humans. Wo examined a vari-
ety of movements generated by human mo-
tion capture data as well as fourteen different
synthetic motion generation algorithms that
wore developed based on human motor pro-
duction theories and computational consider-
ations. In experiments wo displayed motion
clips generated by those 15 different methods
on both a humanoid robot and a computer
graphic character and obtained judgements of
similarity between pairs of movements. Ex-
perimental results reveal that for movements
with obviously different paths as occurred
with two production techniques then, as ex-
pected, hand paths dominated in the per-
ception of similarity. However, for roughly
similar paths as occurred for the other tech-
niques then judgements about fast movements
appeared to bo based on their velocity pro-
file while judgements to slow movements wore
based on a more detailed representation of the
movement.
1. Introduction
The generation of natural appearing motion for hu-
manoid figures is a significant and challenging prob-
lem in humanoid robotics and computer animation.
Wo tackle this problem by drawing on the visual per-
ception of human movement, motor production in
humans, and through the simulation of motion. Wo
have developed fourteen synthetic motion produc-
tion algorithms based on human motor production
Uu
and computational considerations, and have tested
those in two visual perception experiments using mo-
tions displayed using a computer generated figure
and recordings of a humanoid robot. From the re-
sults of observers pairwise similarity judgements wo
reason about what properties of human movement
are salient and speculate as to the possible percep-
tual mechanisms that could explain those results.
In Section 2 wo discuss the visual perception of
human movement and its relevance to motion gen-
eration. In Section 3 wo discuss motion production,
theories of human motor production, describe the
implementation of fourteen synthetic motion algo-
rithms, describe motion capture and explain how wo
produced a collection of motion clips for use as ex-
perimental stimuli. The experiments are described,
along with results in Section 4 and a summary and
conclusions are presented in Section 5.
2. Visual perception of human move-
ment
From the standpoint of Opigonetics robotics it is
useful to start with consideration of the develop-
ment of our ability to perceptually organize and un-
derstand the actions of others. Early studios into
the perception of human movement using movies of
point-lights attached to the limbs of actors mov-
ing in a dark room revealed that infants as young
as 4 months of ago wore sensitive to human move-
ment (Fox and McDaniel, 1982). Further studios,
reviewed by Bortonthal (Bcrtcnthal, 1993), suggest
that infants’ sensitivity to human movement arises
from the same processing constraints as used by
adults, and with the exception of knowledge-based
constraints there is no clear lower-bound on the
ago at which they arc first implemented. Indeed,
studios of 1-yoar old infants have shown a signifi-
cant viewing preference for infants of the same sox