Добавил:
ivanov666
Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз:
Предмет:
Файл:Readings in Robotics Engineering. Учебное пособие
.pdf
3.3. Robot-robot team: cooperative games Video
91
bots are already being used to kill with very little human input. The next step
will be autonomous ‘murderbots’, following orders but ultimately deciding who
to kill on their own. It seems clear to me that this would be completely unethical and dangerous for humanity. We need to be very cautious indeed about
what we ask machines to do.
Which expert has a different view from the others regarding the time-
scale of AI becoming much more autonomous?
Which expert has a similar view to A about the responsibility humans
have for the future of AI?
Which expert agrees with B that the negative aspects of AI far outweigh
the positive aspects?
Which expert seems to disagree with B about how fast humans can
change?
Which expert gives their opinion about what will happen with the same
level of certainty as D?
Find the words in the text:
Take over supersede develop evolve surpass take off go beyond
to replace something older or less advanced
to start doing something that someone else was doing
to suddenly start to be successful
to develop gradually (especially for a biological species)
to grow or improve, or to make something grow or improve
Discussion
Which of the points in the reading do you agree with? Do you see robots
as friends or foes?
https://learnenglishteens.britishcouncil.org/skills/reading/advanced-c1-
reading/robots-friend-or-foe
3.3. Robot-robot team: cooperative games Video
As we saw in the last step, in a human-robot team, the robot knows how
to act because the human controller is telling it what to do. However, in a robotrobot team, the robots must interact and co-operate with each other to decide
what to do next.

Part 3. Working in teams
92
In this video, Dr Michalis Smyrnakis explains how game theory pro-
vides a mathematical framework for robots (in the example he uses UAVs or
Unmanned Aerial Vehicles) to work together to achieve a common goal. We’ll
look at game theory in a bit more detail further.
Transcript
0:05 With the recent advances in technology, there are many cases
where teams of robots should coordinate in order to achieve a common goal.
This includes the classic control aspect of moving the robots from point A to
point B. But there should be also another part that will act as a coordination
mechanism between the robots, in order to take into account what other robots
are doing and how the actions of the other robots influences the whole team
outcome. Game theory provides a mathematical framework in order to achieve
this. The simplest game is matching pennies.
0:58 You have two players, so two persons have a penny in their hand which is 20p actually - and they have to place the coin on the table. If the faces
of the coins are the same, then one player wins. If it’s not the same, then the
other player wins. The simple matching pennies game can be transformed
slightly and work as a collision avoidance mechanism between UAVs. So now,
consider the matching pennies game where two players place a coin on the table, but both of them win if they choose different faces, and both of them lose if
they choose exactly the same face of the coin.
1:53 So, consider two UAVs that are flying in opposite directions, and
they will collide if they keep going to their target. And they have two altitudes
to fly, or a high altitude and low altitude. This gives us a game, like the one that
we described before. If they each choose the same action, both of them try to
fly the low or high altitude, then they cannot accomplish their mission, and they
lose at the game. But if they choose to fly at different altitudes, so one of the
coins is heads, the other is tails, then they can accomplish their mission and
pass with one on top of the other.
2:35 We have a framework of how we can describe the process of
choosing an altitude by game theory, by how the robots decide which altitudes
they will fly. And the answer to this can come from game theoretic learning.
This is usually an iterative process where two players, or in our case the two
UAVs, are flying and change altitudes. Then they learn what the other UAV is
doing, and eventually, they choose different altitudes to fly.

3.4. Game theory
93
3.4. Game theory
Robots of the future will need to work in teams to accomplish tasks.
However, effective teamwork will require coordination among the robots game theory is one way to provide coordination. This step introduces the basics
of game theory.
Game theory for robot teams
Advances in control and automation have made it possible for robot
teams to work together in order to complete a task. When robots work together
in such as way, the action of each robot in the team influences the actions of the
other robots. Therefore, if the robots need to work independently, a coordination mechanism among the robots is needed. Game theory provides such a
mechanism.
Game theory: In game theory, each robot is considered to be a player of a
game and receives rewards dependent on the actions of the whole robotic team.
Reward: A reward is a stimulus used to indicate a desired outcome has
been achieved. A reward for humans is context-dependent, e.g. a gold medal
for winning a race, but for robots is usually arbitrary, e.g. a 0 for no reward, or
a 1 for a reward.
Using game theory, if all robots work in a coordinated way to accomplish a task, each robot will receive a positive reward. Therefore, the goal of the
game is for the team to find a coordinated solution that will maximise the rewards for each robot and the total reward of the whole team.
A simple game for two unmanned air vehicles
Consider two Unmanned Aerial Vehicles (UAVs) flying towards each
other from opposite directions. They can fly either at high or low altitude. The
goal of the two UAVs is to fly at different altitudes in order to avoid collision.
The interaction between the two UAVs can be described by the game
depicted in the table below. In this game:
• One UAV is modelled as a ‘row player’ and the other is a ‘column
player’.
• If both UAVs fail to coordinate by choosing to fly at the same altitude,
they will not receive any reward (i.e. 0).
• Each UAV receives a positive reward (i.e. 1) if they avoid collision by
flying at different altitudes.

Part 3. Working in teams
94
In the game, the rewards of the robot team are represented by a matrix,
as shown in the table below.
Column Player
Fly at high altitude
Fly at low altitude
Row Player
Fly at high
altitude
0,0
1,1
Fly at low
altitude
1,1
0,0
In game theory, a solution in the table with reward 1,1 is known as a
Nash equilibrium.
Nash equilibrium. A Nash equilibrium is a solution to a noncooperative game where each player, knowing the playing strategies of their
opponents, have no incentive to change their own strategy.
Note that once in a Nash equilibrium a player will be worse off by
changing their strategy.
Example: In the UAV example above, once the UAVs are in a Nash
equilibrium with reward 1,1 (representing one UAV flying high, one UAV flying low, and both avoiding a collision), if one UAV changes altitude, this will
result in a collision – i.e. the UAV would be worse off by changing strategy.
Learning algorithms in game theory
Game-theoretic learning algorithms can be used as a coordination mechanism among the robots. These are iterative processes where the same game is
repeatedly played until either coordination is achieved or the maximum number
of iterations is reached.
The learning algorithm follows an iterative procedure – at each iteration,
each robot:
1. Computes a strategy on how to choose an action
2. Selects the best action according to the strategy
3. Checks if coordination is successfully established amongst the joint
action of the team
a) If no, a new iteration starts at step 1 and each robot adjusts and updates their strategy
b) If yes, the learning algorithm terminates.
The general procedure of game-theoretic learning algorithms can be represented by the following figure.

3.5. Team work and swarms Video
95
The basic principle behind these algorithms is that robots use the history
of observed actions in order to predict the other robots’ strategy and then
choose an action based on their prediction. The key result is that robots successfully learn to play the game.
3.5. Team work and swarms Video
Swarm robotics provides another mechanism for the coordination of
multi-robot systems. In robot swarms, a desired collective behavior emerges
from the interactions between the robots and interactions of robots with the
environment.
In this video, Dr Roderich Gross introduces us to the e-puck and explains how, with very little brain power, these robots can work together to successfully achieve a complex common goal.
Transcript
0:09 This is a simple robot called e-puck that we have been using for a
while. And it has two wheels, a left wheel and a right wheel, so it can move
forward, turn around, turn on the spot, and so on. It also has a camera right here

Part 3. Working in teams
96
in the front, and this camera is useful for the robot to understand where are the
other robots, if there are other objects of interest in its world. What you’re go-
ing to see now is how a bunch of these robots are solving a task together. They
are clustering a group of red objects and bringing them all into one spot. So this
is a bit like collecting litter, right?
0:51 The fascinating thing about this is that these robots are very, very
simple. So, what information do they actually have about the task? They have
this camera, and using this camera they scan, just in front of themselves,
whether there is an object in front of them that they have to cluster, another
robot, or simply nothing. So there’s just three different situations, and they only
discriminate between these situations. And depending on whether they see another object, another robot, or nothing, they behave in different ways. So they
have preprogrammed motion primitives. For example, one of these could be
turn towards the left, and so on.
1:31 And they simply map the input that they perceive through their line
of sight sensor directly onto their action, which is this motion primitive. As a
consequence, these robots do not require the ability to store information during
run time, and they also do not even need to compute like plus, minus and so on,
because the controller is really just mapping from the input to the output. Biological systems are really good at solving tasks under a lot of constraints. They
need to solve tasks with very little energy, for example. And by studying robots
that are inspired by biological systems, we can potentially build very simple
robots that yet accomplish complex tasks.
See also
1. TED TALK: Robots that fly ... and cooperate
In his lab at Penn, Vijay Kumar and his team build flying quadrotors,
small, agile robots that swarm, sense each other, and form ad hoc teams —- for
construction, surveying disasters and far more.
https://www.youtube.com/watch?v=4ErEBkj_3PY
2. Occlusion-Based Cooperative Transport with a Swarm of Miniature
Mobile Robots
In this paper, Roderich Gross and colleagues propose a strategy for
transporting a large object to a goal using a large number of mobile robots that
are significantly smaller than the object.
https://ieeexplore.ieee.org/document/7055285

3.6. How future robotic teams might work together Video
97
3. Supervisory control theory applied to swarm robotics
In this paper, Roderich Gross and colleagues propose the application of
supervisory control theory (SCT), a framework for formally synthesising controllers, to the domain of swarm robotics.
https://link.springer.com/article/10.1007/s11721-016-0119-0
The future for robot teams
What are the possibilities for robot teams? Our experts discuss where
you might encounter robot teams in the future and how they will work together
in different environments.
3.6. How future robotic teams might work together Video
Sheffield’s experts share what the future will look like for team working
robots.
First up, Dr Jonathan Aitken gives us a glimpse into the factories of the
future, explaining how the Kuka iiwa robot arm will work alongside workers on
the factory floor. We’ll then hear from Professor Noel Sharkey about robot
surgeons and from Dr Roderich Gross about swarms of exploratory robots.
Transcript
0:13 This is KUKA’s lightweight arm. It’s called the iiwa, which is their
next generation of industrial robotic arm. This is the academic version, which is
capable of carrying roughly seven kilos and has a reach of roughly 800 millimetres. We’ve got it equipped here with a Schunk Dextrous Hand, so a threefingered dexterous hand, which also has tactile portions on the fingertips which
let you know how hard it’s gripping something. The robot arm itself has been
developed for future use in factories, particularly looking at co-working scenarios where a member of the factory floor staff will be working with a robot. That
robot will be doing the dull, repetitive parts of their everyday tasks.
0:51 It means they can be getting on with their normal work doing the
more dexterous parts, the more highly skilled parts, while the robot does the
parts that will actually take their time. One of the ways that we can actually
show how this robot will actually do this task is through something called compliance. Compliance means that we’re effectively able to back-drive the motors
given a force someone applies. That means I can actually begin to teach it a

Part 3. Working in teams
98
various set of movements that it has to make. So this could be, for example, an
everyday pick-and-place task which I would have to manually do myself.
1:20 But I can actually do this with a robot, and actually get the robot to
perform this task itself. And what that means is, because I’ve taught it the task,
it can actually replay that task endlessly, 24/7, doing exactly the same job and
the same movements over and over again. And that frees me up to do the other
work that I need to do on the line. We’re able to do this because the motors are
back-drivable. The motors will actually allow the arm to move, but then keep it
in position.
1:44 This is one of the ways that actually makes this arm safer, because
traditional robot arms that you see in industry these days on factory lines typically
need to be caged in. That’s for the safety of any operators who need to be outside
that cage, because the arm will not detect them. This back-drivable compliant
nature of this arm means we can actually closely interact with the arm and it is
actually then safe to use. One of the other aspects of this arm that we have is the
multiple degrees of freedom that we can actually see within the arm itself.
2:11 So for example, as I begin to rotate the arm, you can see all the joints
of the arm are moving, but I’m actually remaining in a fixed position in space.
What this means, and what this provides us, is a way of actually planning for
motion of the arm to avoid problems and issues where particular joints may come
into collision with each other, and actually conflict and cause a collision. By careful planning and careful use of this redundancy, we can actually get around these
problems. Of course, one of the big areas where there’s going to be major break-
throughs as well, I think, is in robot surgery and robots in medicine.
2:45 Because currently, the robot surgeon - it’s sort of a remote controlled
thing called the daVinci Si, which is mostly used - there’s a couple of others
around as well. And so the surgeon has her or his head in a console, working with
levers, but you can set it up so that somebody has shaky hands or makes big
movements and it only moves a small way. It’s basically keyhole surgery.
3:08 In medicine, it’s also the idea of remote surgery, so that this idea
that the doctors are in there with the console, but you could make it remote so
that for instance, I can see in the future, and there’s been some research on it,
putting one of these small surgeon things on ambulances. So you’ve got a
pileup on the motorway ‒ most people die coming back to the hospital, that’s
where you die – or some sort of rescues, emergency, storms or anything like
that. So you send these little surgeons out, and one doctor with the help of as-

3.7. The possibilities for robotic teams
99
sistants on the ground could do temporary operations on a number of people,
make them secure to bring them back to the hospital.
3:48 And the US military were working on that for battlefield, but they
cancelled programme, unfortunately. It would have been good for all of us.
There has been an operation already between Buenos Aires and Paris. So a surgeon in Paris was operating on someone in Buenos Aires. They have to have a
dedicated line, of course, because if it breaks, you’ve really had it. So I could
see that that would be a very good use as well, because lots of places in the
world don’t have the possibility of having a surgeon.
4:15 And if you could just fly in one of these, when it’s needed in some
little village somewhere, or in some war zone, you could fly it in and the surgeons
can operate from safety at home and not have to go travelling or whatever. By
working in teams, robots can accomplish a mission that is impossible for a single
robot alone. So, for example, they could push around objects that are a very bulky
and heavy that a single robot is not capable of moving around. And there’s many
more examples where robots can benefit from collaboration and in particular
where the environment gets physically manipulated. Other areas for swarms of
robots are inaccessible spaces like the deep sea, or space, even other planets.
5:05 So if you send a team of robots there, they can explore the environment, they can explore much more environment than a single robot, and
they also are inherently redundant. If a single robot breaks down, it’s not that
the entire mission has failed. So having swarms of robots brings us flexibility
and robustness, and therefore in principle, you can apply them to virtually any
type of environment.
3.7. The possibilities for robotic teams
Robot teams will be essential for solving certain complex tasks. So,
where might you encounter robot teams in the future? In this step, we look at
some of the future applications for robotic teams and discuss how they will
work together in different environments.
Factories and warehouses
Robots can be asked to fetch and bring items to the counter using a
computer-based order system. If an item is too bulky, two robots will automatically cooperate to fetch and deliver the item.

Part 3. Working in teams
100
A manufacturing process may require more complex coordination.
A human supervisor will issue a command to a team of mobile robots on the
factory floor and the team will need to jointly plan and negotiate who is going
to do what, based on current individual capabilities. Executing the task will
require constant re-planning and monitoring so that the work progresses or
adapts from the agreed joint plan.
Search and rescue
The team will first be provided with information about the area they
need to search and will be “briefed” about the nature of their responsibilities.
This might be gathering information or delivering emergency supplies.
From here onwards, the robots need to create a joint plan of the initial
steps and search policy that they will follow during the actual search activity.
Executing the mission will require constant updating and sharing of a combined
environmental model.
Agriculture
Robot teams work together to harvest fruits or collect vegetables from
the ground in a large production field with a single human supervisor. The robots need to coordinate their team movements and minimise their total travel
path to reduce costs and complete the harvest within the shortest time period.
While collecting the harvest, they will update a shared model of overall progress using their localisation and perception systems. They will inform the human supervisor of any practical problems they face in terms of robotic hardware problems or difficult to access terrain due to water and other environmental hazards.
Other types of robot teams, such as drones may help the farmer to
achieve the highest yields on his fields by imaging the land, analysing the images and advising the farmer on soil treatment methods.
Infrastructure
Robot teams may work together to clean the streets from litter on Friday
to Saturday night so that streets are clean by the morning.
Teams of flying robots can help to survey a large construction site for
information of work progress and help to optimise management.
Buildings and structures such as bridges and masts can be surveyed for
their repair needs. Teams of robots can execute survey tasks more quickly and
accurately and the robots can carry different sensing capabilities (3D geometric
mapping, thermal mapping, X-rays imaging, ultrasonic crack detection, etc.)
Соседние файлы в предмете [НЕСОРТИРОВАННОЕ]
