Active Perception For Robot Manipulation
Active Perception For Robot Manipulation
Name
Course
Institution
Date
2
Introduction
Active perception can be defined as the increase of information intake from the source or flow of
data by a selection of certain behaviors that can help achieve this intended goal which is the
production of a particular environment [1,3]. This means that for a better understanding of what
is around a certain entity, there is a need to move around and explore it by sampling, observing,
feeling, and analyzing what each aspect means all based-on behavior. Active vision as a specific
point of view of active perception can be defined as a study of modeling and controlling
strategies for perception [2]. Unlike animals with a vision which can control their body and
sensory organs like eyes, body, and heads to explore their surroundings and process what to
make of it to make their decisions, the concept has been applied over the years in the field of
artificial intelligence specifically robotics to help machines perceive more of the environment by
use of cameras that are fixed on to them and can rotate also the use of partial sensors have been
used to increase the experience and make them more perceptive of the surrounding and collect
The main goal in these types of necessary adjustments and improvement is to ensure that like
animals, robots can see the environment from different viewpoints which can, in turn, provide a
clear and rich visual experience that will be used to learn, analyze and make a decision on it [3].
This literature review examines how active perception is applied in the robotics field and tries to
highlight the advancements made in the field to make robots behave and perceive the
environment more like animals and how it has increased productivity [2].
3
History of Robotics
Although mechanical devices have been present since even in the times of the Egyptians, the
inventions in the field of robotics saw light in the 1950s. A scientist inventor, George C. Devol
from Louisville, Kentucky made and patented a reprogrammable manipulator and named it
“Unimate” from “Universal automation” [1]. Although he did not succeed in selling the
invention for a decade, Joseph Angleberger an engineer and businessman acquired the invention
and modified it into an industrial robot which was under the company “Unimation” which was a
Over the years, research in the field increased and more inventions have been made ever since.
Academia made “Shakey” in 1958 which was an improvement of the “Unimate”. Most of the
robotic inventions at this time were able to do a limited task which was expensive with little
output, with the rising of other fields such as data science and the processing of data that comes
with it, the inspiration of sensory equipment like cameras and other sensors seemed appropriate
as it would make them more efficient in the daily working of the robots, these all began in the
late 1970s and still ongoing as with each improvement the robots can gain active vision.
Leonardo DaVinci developed drawings for a Germanic knight-shaped item that could mimic the
exact moving patterns of a man, however, it is not clear if he did make any fabrications [3,6].
Hans Bullmann, a German academic, creates humanoid androids which plays music around
1525. In 1737, French engineer Jacques Vaucanson created many artificial creatures, the far
more notable of which was a ducks that could do all the processes of a living duck except
breathing. In 1801, Joseph Jacquard, a French inventor, creates an automated loom operated by
punch cards, which marks a shift in the intended purpose of automata devices from entertaining
to industry. Babbage's theories and developments improved with his steam engines, as well as
4
the subsequent presence of George Boole's Boolean algebra, laid the groundwork for the
implementation of the newly emerging algorithmic scientific fields in the generation of different
computational models also with goal of building true nearly human skills between both the 20th
and 40th centuries. The tweentienth century is observed to have had a high impact in the
dvelopment of the history of robots and robotics. It was time when many researchers and
scientists appeared to have been so intrigued by the concept of robotics and so it rose as fast as
whirlwind. Karel Capek invented the term "robot" for the first time (what means hard work). For
his part, Isaac Asimov created the three rules of robotics in his book "Runaround," which are still
true today. Along with the emergence of related fields like as computer science and artificial
intelligence, the issue of developing automated machines/tools continues [3]. Also, the very first
controllable adjustable robotic arms was created by engineers George Devol and Joe Engleberger
in 1954. A few years after that , that would be the first robotic to be employed in industry,
specifically in the automobile industry, and the first robot to be mass-produced for these reasons.
Ever since and until today, automation has followed humankind's big events, adding to technical
advancement and new information, as well as assisting man in tough work [4,6]. Robotics were
there in the race to the moon, they are all on Mars, they have assisted in the exploration of an
ocean depths, and that they have assisted in the examination of the damages done by a nuclear
plant disaster, when a person would simply die attempting to accomplish it.
Although most parts of robotics have managed to make advances and can perform given tasks at
an improved rate than that of their predecessors, active vision is still a bigger problem. Things
like a restricted field of view, occlusion, and limited resolution all affect a robot’s ability to sense
the surrounding environment and unlike animals, this is the most important of all senses it
5
requires making its need more crucial [3,1]. In active inference, we say that the robot’s
intelligence needs to maintain a generative model where it can act according to the given
instance to reduce and avoid surprises. But due to the current level of technological
advancements, the problem of robot manipulation still presents a big challenge when working
With the need to ensure that robots are more of a generative model in action, some recent works
have illustrated how active vision as a form of perception emerges from active inference by
running a lot of simulations [1]. However, these approaches define the generative model in an
upfront manner which is small and discrete and not in a larger scale. This leads to the need of
robotic manipulator to aid robots in active perception. A robot manipulator is a very essential
interacting with the environment [3]. Some of the known manipulators known are polar,
Cartesian, cylindrical, articulated, SCARA and delta configuration which all do the main job of
manipulating the materials in surrounding environment without initiating direct physical contact
with operators and hence help in the active perception for robots.
In robotics, an autonomous agent is defined as an intelligent agent which operates on the user’s
interphases without causing any interference to the user at any instance. However, an intelligent
agent is defined as a software which performs a given set of operations on behalf of the user or
perhaps on behalf of another separate program with some level of independence by following the
user’s objectives, instructions, and goals [5,2]. These agents may include systems in a
technological or surrounding habitat which perceives all or even some state of that surroundings
then reacts around it to forward its very own goal is an example of an agents. Drivers shape one
6
such objective. The agency modifies what it perceived by acting to modify an element of the
The activity, or lack thereof, is the essential distinction with an active vision technology and
other perceiving methods. While both structures have choice elements, just the proactive method
combines both outside (through motors) and interior (by system parameters) dynamically
modulation formats to the total agent's behavior [4]. Imagine the following scenario: we
successfully developed filters to detect a certain object, such as a sword, from photos using
cutting-edge machine learning approach. Take into account the fact that our filter has a 90%
successfulness. That might be a ground-breaking conclusion, but it may not be as exciting for a
robotic system that is functioning [2]. The core of proactive perceiving is to select a goal
predicated on a present world view and then put into motion the behaviors that will help you
attain it. In those other terms, proactive perceiving is purposeful because it must synthesize
information and movement in order to attain a goal [6]. To build a concept, one must have a
Conclusion
From the history of robotics, it is clear that robotics has taken huge strands and even more in the
current modern world. There are still many more sophisticated innovations that the world is yet
to witness. However, we have also observed that robots still need a lot of work to be done on
them. This will ensure that like animals, robots can see the environment [4] from different
viewpoints which can, in turn, provide a clear and rich visual experience that will be used to
Bibliography
[1]
[2]
[3]
[4]
2018. http://introtocommopensource.ridgewater.edu/ModuleII/ModIISect2.html
[5]
Q. Le, A. Saxena, and A. Ng, “Active Perception: Interactive Manipulation for Improving Object
http://ai.stanford.edu/~quocle/activevision.pdf
[6]
2022).