Information Technology Reference
In-Depth Information
Recently, several types of high-speed vision in which the sampling rate is more
than 1 kHz have been developed [4, 6]. In the high-speed vision system, not only
sensing but also processing is achieved at high-speed. It is the main difference of
the high-speed vision from a high-speed camera which is used for recording the
video in slow motion. In order to achieve rapid processing, mostly a type of parallel
processing architecture are used.
There are several advantages of high-speed vision in manipulation tasks. First,
the high-speed vision is useful to recognize an object moving at high-speed. For a
high-speed vision system, the object moving at high-speed seems to be stopping.
For this reason, a complex prediction process is not needed. Because the transition
during one frame is small, the scanning and exploration process in the image plane
become also useless. The high-speed vision is also useful to recognize a flexible
transformable object such as rope or cloths. Secondly, a high-speed vision at a rate
of 1 kHz can be included to a feedback control loop directly. In robot manipula-
tion systems, the sampling rates of other sensors such as an optical encoder and a
strain gage are about 1 kHz. A high-speed vision system can be used with the other
sensors simultaneously. Also, because the delay is small, a high-gain visual feed-
back can be achieved. These advantages show that a high-speed vision improves
the performance of visual servoing. As a result, a visual control only based on the
appearance without the precise calibration can be achieved.
Our research group have developed a manipulation system with high-speed vi-
sion, and by using the system several types of high-speed manipulation have been
achieved: dynamic regrasping [1], stick spinning [3], and rope knotting [11]. In this
chapter, we propose a high-speed batting task [9, 10] and a tool manipulation task
[5] as examples of the visual feedback control using high-speed vision.
3.2
High-speed Batting
A human batting task in the baseball game is a task which needs high-level visual
control ability. In general, it takes a human more than 0
.
2stoactbasedonvisual
information, and it takes a ball only 0
1 s that to reach a catcher's mitt. For this
reason, a human must predict the ball trajectory using some other information such
as the pitcher's motion. However, it is difficult to predict it perfectly, and he often
misses. If we use a high-speed vision, we can realize a more reliable batting system.
We proposed a hybrid trajectory generator as a motion strategy for a high-speed
robot system and achieved the batting task [9, 10]. This trajectory consists of both a
swing motion and a tracking motion.
.
3.2.1
Hybrid Trajectory Generator
The batting motion consists of two motions as shown in Figure 3.1: one is the swing
motion, and the other is the hitting motion. The swing motion is necessary to hit a
ball hard, and it should be controlled so that its speed approaches the maximum. In
order to achieve a smooth high-speed motion, this motion should be represented as
Search WWH ::
Custom Search