Page 168 - Human Inspired Dexterity in Robotic Manipulation

166 Human Inspired Dexterity in Robotic Manipulation

such information in advance is difficult because of the great diversity of the
types of objects that are present in a typical daily living space. Another way to
obtain such information is to measure the information in real time. For the
perception of a grasped object, various kinds of sensors have been used to
date, such as visual, tactile, force sensors, and encoders. Above all, visual sen-
sors have recently been used most frequently as the most popular type of
sensor for measuring an object’s position and orientation because they
can perform contact-free measurements over a wide range. To date, visual
sensors and sensing techniques have been evolving and becoming increas-
ingly easy to use even for users with no related expert knowledge. However,
several problematic limitations still exist with regard to time, such as low-
sampling rates and heavy computational costs for image processing. The
average sampling rates of easily available cameras and RGB-D sensors still
typically range from approximately 30–120 Hz, with the exception of
expensive high-speed cameras. Also, the sampling rates of visual sensors have
been witnessing continuous improvement. These rates, however, are still
insufficient for direct sensory feedback control of a hand-arm system because
the servo loop used to control the joint torques of a robotic system generally
requires a sampling period on the order of a few milliseconds. Moreover, the
rates depend on image resolution, meaning that wide-range and highly
detailed images decrease the sampling rate. In addition, some time is inev-
itably required for image processing calculations and communication, and
this time cost is also problematic. Heavy calculation processes can cause
unstable behavior of a robotic arm even when the sampling rate of the visual
sensors is as high as that of the robotic arm. In other words, the considerable
time delays caused by data processing and communication directly affect the
severity of the overall time delay. The conventional approaches for coping
with problems related to time delays are state-estimation techniques such as
Kalman filters [1], AR models [2], nonlinear observers [3], and multirate
controls [4, 5]. These methods are based on linear approximation tech-
niques. Robots that do not interact with movable objects such as mobile
robots are well suited to these techniques. However, a linear approximation
is not appropriate for the overall dynamics of an object-manipulation system
because such a system includes many geometrical and physical constraints
with regard to objects. In fact, studies of these methods have mostly focused
on reaching or moving tasks executed by the end effector, and the effects of
time delays on dexterous object manipulation have not yet been discussed.
Recently, we have proposed a stable-object grasping and manipulation
method for an arbitrary polyhedral object using a multifingered hand-arm