Download How Many Robots Does it Take to Screw in a Lightbulb?

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Incomplete Nature wikipedia , lookup

Behaviorism wikipedia , lookup

Philosophy of artificial intelligence wikipedia , lookup

Existential risk from artificial general intelligence wikipedia , lookup

The City and the Stars wikipedia , lookup

Kevin Warwick wikipedia , lookup

Embodied cognitive science wikipedia , lookup

Adaptive collaborative control wikipedia , lookup

Index of robotics articles wikipedia , lookup

Robot wikipedia , lookup

Self-reconfiguring modular robot wikipedia , lookup

Robotics wikipedia , lookup

Robotic automation software wikipedia , lookup

Ethics of artificial intelligence wikipedia , lookup

Transcript
Righettini Page |1
David Righettini
WRIT 340
Prof. Ramsey
11/2/2011
How Many Robots Does it Take to Screw in a Lightbulb?
Emergent Behavior in Swarm Robot Systems and their Potential Use to Humanity
Source: http://singularityhub.com/2009/06/05/swarm-robotics-beware-the-swarm/
Introduction: Greater than the Sum of their Parts
Many people have the preconception of a robot being something like what you would see in a
Star Wars movie - a single mesh of circuitry and wires that acts in a relatively independent fashion.
However, an emerging trend in modern robotics is the use of swarm-type robots. As the name suggests,
this involves a large group of robots that work together in order to accomplish some task that a single
robot would struggle to complete. Just like a single computer is limited when it can't talk to other
computers on the internet, the effectiveness of a single robot is limited. Swarm robots attempt to
overcome this disadvantage by mimicking the interactive behavior of animals and humans, creating a
group that can solve meaningful problems in daily life. This has some fascinating results that have
spurred researchers to investigate an interesting phenomenon called emergent behavior. When
multiple robots interact with simple rule sets or instructions sometimes they display an ‘intelligence’
that was not explicitly programmed into the robot. These general instructions end up being robust and
powerful but somewhat unpredictable. Despite the fact that it is hard to guarantee the exact behavior
of the swarm, it is quite possible that this style of robotics can be more effective than using lone robots
if emergent behavior is thoroughly developed. Once robot swarms are reliable, they will be able to
improve the quality of human life.
Righettini Page |2
To Swarm or Not to Swarm?
Traditionally, robots have been single entities that act on their own. Take Boston Robotics'
Bigdog for example.[1] It is a robot designed to carry equipment and finds applications for soldiers on the
battlefield. However, as with most single robot systems it finds a niche use with the military, not the
general public. Multi robot systems have the potential to be superior to single robot systems. Maze
navigating robots offer a practical example, as they have to overcome the problem of navigation - a
critical aspect of both man and machine. Maze navigating robots are very popular in academics since the
problem is well defined and has enough complications to make it challenging. If a developer decides to
create a single robot to map the maze, all he/she would have to do is program it to search the entire
maze and take measurements. Ideally, searching the maze would be quicker if there were more than
one robot. However, this adds another layer of complexity to the challenge, specifically, how do the
robots talk with each other? This makes swarm robots difficult to program and hard to implement. But
there are many rewards for successfully taking advantage of multi-robot systems. There are generally a
large number of robots working together, so it makes little difference if a few of them break down or
become inactive. Since each robot uses the same set of rules and instructions, losing a few does not
really change the overall effectiveness of the system. For example, Lynne Parker has been developing
the ALLIANCE software architecture as a system that allows for robots to be taken offline and still
permits the swarm to accomplish its objective.[2] Conversely, if the system only consists of a single robot,
the robot will fail its objective if a critical part breaks. Additionally, swarm robots can cover a much
larger area than a single robot could ever hope to accomplish. However these swarm robots are also
difficult to program and can get pretty expensive if many are built for a project. Besides these factors,
there is also a level of unpredictability that goes with programming multi robot systems.
Emergent Behavior: Friend and Foe
The programming difficulty comes from a
side effect of multi robot systems called emergent
behavior. Both a help and a hindrance, emergent
behavior is the defining feature of swarm robotics
and it is one of the keys that unlocks the potential of
swarm robots. It is really hard to come up with a
concrete definition for emergent behavior. It is like
Figure 1: The self-organization of these birds is an example
of emergent behavior. Source:
http://www.patternsinnature.org/Book/EmergentPatterns
.html
Righettini Page |3
trying to explain common sense. In the simplest form, emergent behavior could be thought of as getting
unexpected behavior from simple instructions. For example, if a programmer develops two robots that
only have instructions to move towards a ball, we may find that they could effectively play soccer with
one another. They were not explicitly programmed to play soccer, with all its details, yet they managed
to perform adequately in that capacity. However, it is overly simplistic to think that everything
unexpected is emergent behavior. Instead it "is structured behavior that is apparent at one level of the
system (the observer’s point of view) and not apparent at another (the controller’s point of view)".[3] In
order to take advantage of this sort of methodology, programmers have to let go of their tendency to
take every little bit of logic into account and, instead, they need to find a few basic rules that cover the
scenario and hope that the robots work effectively. There is always uncertainty since it is difficult to
determine exactly what will happen beforehand, but the result is a simple and effective methodology
for using swarm robots.
Learning from the Ant
The way robots communicate produces emergent behavior, and many systems can exploit this
inherent behavior simply by using certain communication schemes. Even distributed systems that do not
try to take advantage of emergent behavior find that their systems still exhibit such behavior[4]. There
are many methods available. Some robots use different colored lights to communicate. Each robot has
a camera and can determine the state of the others by analyzing the color. This is a cost effective
solution, but it is limited by line of sight. If the robot does not see the light it will be unable to
communicate with other robots. Two researchers named Brandoff and Sayama have seen success by
using RFID tags that can store data. These tags use radio signals to store data. As robots pass by, they
can send and receive data from the chips and get a general idea of what the other robots have
discovered.[5] This allows the robot to make smarter decisions, because it has the knowledge of multiple
robots. Development of new communication methods is critical for the advancement of swarm robotics
since the robots will simply get in each other's way otherwise. However, it is the truly simple and generic
type of communication patterns that tend to be the most effective and widely applicable. One of the
most intrinsic algorithms in swarm robotics was developed by studying patterns in nature, specifically
ants. Observation of ants reveals that they normally travel to and from their food in a straight line. But
these ants don't send explicit messages to each other, so how do they all know where the food is?
Essentially, the ants will wander individually until one of them finds food. As they move, they drop
pheromones on the ground that other ants follow. Since all ants take the shortest path back to the nest
Righettini Page |4
from the food, that trail will eventually have the highest concentration of pheromones – meaning that it
will be instantly followed by the ants[6] (see figure 2). Many core search algorithms for swarm robotics
use the ant colony principle as a basic
component because it is very effective for
searching large areas effectively. However,
since robots do not drop pheromones, and
dropping bread crumbs is a bit impractical,
Figure 2: Example of ant pathfinding. Source:
http://ljs.academicdirect.org/A09/053_072.htm
developers are looking into new ways to
exchange information between robots in order
to take advantage of the ant colony algorithm
and other similar computer science
methodologies.
Divide and Conquer: Using the Swarm
Simpler algorithms have changed the shape of robotics allowing many robots to interact and
accomplish tasks together. No one has really found a mainstream or marketable application for robots
(anyone remember the Roomba?), but swarm robots have the potential to truly affect the lives of
people in a beneficial manner. However, there are still quite a few valuable and useful large scale
applications for robots. Many of these applications involve exploration, navigation or searching.
Consider NASA: it had planned to use swarms of robots to search asteroid belts.[7] Swarms are perfect
for covering large areas. Another possible application for swarms is firefighting. Researchers, Wang et
al, have simulated scenarios where large groups of robots locate and track fires on multiple fronts using
a modified version of the ant colony algorithm.[8] Such applications indicate a bright future for swarm
type robots, as they are showing potential to effectively solve complex tasks. Eventually, many of the
dangerous or inconvenient tasks that society is faced with could be solved using a group of robots.
A Look to the Future: Robots Will Not Take Over the World
With a little imagination, it is not too farfetched to see a point in time where robots are used in
day to day life. Take transportation as an example. In 2005, the DARPA Grand Challenge proved that
robotic vehicles had the ability to successfully traverse over 100 miles of the Mojave Desert using
nothing more than GPS localization and a net of sensors[9]. Using these robotic vehicles and connecting
them in a sort of multi-robot system could be the key to solving transportation issues in large cities. An
Righettini Page |5
effective swarm robot system could take just about any activity that humans perform as a team and
accomplish their goal in a consistent manner. Imagine teams of robotic construction workers that
receive uploaded schematics and proceed to assemble a building. These robots could assemble a
building with a higher level of precision and speed while reducing injury and toil of human workers.
Granted, this sort of technology will probably not come to
pass for at least a few decades. Nonetheless, it should be
kept in mind that swarm robots have great potential to
shape the choices of way we do things in the activities of
daily living. It is evident that many robots working
together can accomplish much more than robots working
alone(see figure 3). With some more work and research,
robots can be taken out of the realm of science fiction and
fantasy into the modern world, where they can provide
Figure 3: Swarm robots working together. Source:
http://www.swarm-bots.org/
utility to humankind.
References:
[1] Raibert, Marc; Blankespoor, Kevin; Nelson, Gabriel; Playter, Rob. " BigDog, the Rough-Terrain Quaduped Robot
" Boston Dynamics. WEB: http://www.bostondynamics.com/img/BigDog_IFAC_Apr-8-2008.pdf
[2] Parker, Lynne E. ALLIANCE: "An Architecture for Fault Tolerant Multirobot Cooperation" IEEE TRANSACTIONS
ON ROBOTICS AND AUTOMATION, VOL. 14, NO. 2, APRIL 1998
[3] Nicolescu, Monica. "Topics: Introduction to Robotics" PowerPoint. University of Neveda. Internet FTP:
www.cse.unr.edu/~monica/Courses/CS491-691/Lectures/lecture11.ppt
[4] Moshirpour, M.; Mousavi, A.; Far, B.H.; , "A Technique and a Tool to Detect Emergent Behavior of Distributed
Systems Using Scenario-Based Specifications," Tools with Artificial Intelligence (ICTAI), 2010 22nd IEEE International
Conference on , vol.1, no., pp.153-159, 27-29 Oct. 2010
[5] Brandoff, J.; Sayama, H.; , "Cultural transmission in robotic swarms through RFID cards," Artificial Life, 2009.
ALife '09. IEEE Symposium on , vol., no., pp.171-178, March 3 2009-April 2 2009
[6] Macura, Wiktor K. "Ant Colony Algorithm." From MathWorld--A Wolfram Web Resource, created by Eric W.
Weisstein. http://mathworld.wolfram.com/AntColonyAlgorithm.html
[7]Wang, D.; Kwok, N.M.; Fang, G.; Xiuping Jia; Li, F.; , "Ants Based Control of Swarm Robots for Bushfire Fighting,"
Artificial Intelligence and Computational Intelligence (AICI), 2010 International Conference on , vol.1, no., pp.528532, 23-24 Oct. 2010
Righettini Page |6
[8] Rouff, C.; Vanderbilt, A.; Truskowski, W.; Rash, J.; Hinchey, M.; , "Verification of NASA emergent systems,"
Engineering Complex Computer Systems, 2004. Proceedings. Ninth IEEE International Conference on , vol., no., pp.
231- 238, 14-16 April 2004
[9] Orenstein, David. "Stanford team's win in robot car race nets $2 million prize" Stanford Report, October 11,
2005