Teaching by touching:
an intuitive method for development of humanoid
robot motions
Fabio Dalla Libera∗, Takashi Minato†, Ian Fasel†, Hiroshi Ishiguro†‡, Emanuele Menegatti∗ and Enrico Pagello∗
∗ Intelligent Autonomous Systems Laboratory, Department of Information Engineering (DEI),
Faculty of Engineering, University of Padua, Via Gradenigo 6/a, I-35131 Padova, Italy
† ERATO, Japan Science and Technology Agency,
Osaka University, Suita, Osaka, 565-0871, Japan
‡ Department of Adaptive Machine Systems, Osaka University, Suita, Osaka, 565-0871 Japan
Abstract— This paper investigates touching as a natural way
for humans to communicate with robots. In particular we
developed a system to edit motions of a small humanoid robot by
touching its body parts. This interface has two purposes: it allows
the user to develop robot motions in a very intuitive way, and it
allows us to collect data useful for studying the characteristics
of touching as a means of communication. Experimental results
confirm the interface’s ease of use for inexpert users, and analysis
of the data collected during human-robot teaching episodes has
yielded several useful insights.
I. INTRODUCTION
In order for robots to become truly integrated into everyday
life, it will be necessary for humans to be able to interact
with them in a natural and intuitive way. This consideration
has recently lead to many different studies in human-robot
interaction with the aim of finding natural ways by which
humans can communicate with robots (e.g., [1]–[4]).
Abstractly, we regard communication as a process by which
a sender encodes a concept into a format suitable for trans-
mission though a medium, and sends this information to a
receiver, who then reconstructs (or decodes) it. We can divide
human-to-human communication roughly into verbal (when
concepts are encoded in the form of words) or non-verbal
(see for instance [5]). Non-verbal communication can then be
broken down further according to the transmission channel
used, such as communication via vision, smell (a study on
this communication medium is provided by [6]), or touch.
There are a great variety of studies dealing with characteristics
of visual communication, for example recognition of human
gestures [2], analysis of how a robot’s aspect affects human-
robot interaction [7], and knowledge transfer through visual
task recognition, as is in Kuniyoshi and Inoue’s [8] work in
which a robot observes humans performing a task and rec-
ognizes various actions, from which it constructs a high-level,
abstract plan. However touch as a communication medium has
received considerably less attention.
Touch is an important method of communication employed
by humans, particularly in teaching. Even at the earliest ages,
touching behaviors have been found to be a very important
element of interactions between humans and preschoolers [9].
At older ages, touch is frequently used in the teaching of
sports or dance [10], for instance by instructors correcting a
learner’s posture or motion. Touch is particularly appealing as
an intuitive method for humans to teach robots, and has been
employed to program robot arms, for example, by Voyles and
Khosla in [3] and, more recently, by Grunwal, Schreiber, Albu-
Scha¨ffer and Hirzinger in [4].
In this paper, we investigate the effectiveness of touching as
a mechanism for transferring knowledge about the body from
a human to a small humanoid robot. Small humanoid robots
are quite popular and are becoming increasingly available
at relatively low cost. However teaching a new motion to a
humanoid robot is currently a time consuming task, because
the standard method is through the use of motion editors
which require the user to set the position of each joint in each
“keyframe” (as illustrated in [11]). Although other techniques,
such as motion capture and retargetting [12], can be employed,
these methods are still somewhat cumbersome, and require the
human teacher to learn specialized techniques.
Our goal is to create a method by which humans can
intuitively instruct a robot without any special training. We
therefore have developed a method for humans to teach robot
motions through an “observe and correct” cycle, similar to a
human dance or sports instruction. In each teaching episode,
the human teacher watches the robot perform a motion,
observes what is wrong or could be improved, and touches
the robot’s body parts to instruct the robot how to modify the
motion. For example, the teacher could watch a kicking motion
and notice that the right leg should rise higher in a specific
moment. S/he can then touch the leg from the back and push
it upward to express how the motion should be modified. The
robot then repeats the behavior with the modifications, and
the cycle can be repeated several times until the movement is
satisfactory to the teacher.
The teacher’s touching actions are a method of encoding
and transmitting their internal image of what the robot postures
should be. To make communication successful, the robot must
then interpret the meaning of these touches in terms of adjusted