UWE Bristol Engineering showcase 2015

Lee Paplauskas
BEng Robotics (Hons)
Project Supervisor
Prof. Alan Winfield
Interaction between Multiple Robots with Self-Simulation
The Future!
Robots, and Artificial Intelligence are become much more prevalent
in our culture. Whether something as basic as an algorithm to
recommend TV shows, or a fridge with twitter to let you know
when you're out of milk, right through to a car which can drive
itself, at the push of a button, its all there.
As a result of this, the machines, or robots, we make, need to be
made safer. There is only so much that can be done to make their
processes safer, using current methods, such as putting proximity
sensors on, or limiting their speed. What if, instead of limiting
them, to make them safer, we gave them a way of predicting if
something they were about to do was dangerous? Why stop at that
though? Why not even go as far as to make the machines protect
us from dangers, outright?
The Three Laws of Robotics
Isaac Asimov wrote, in his short stories about “Positronic” or
emotional robotics, of “The Three Laws of Robotics”. These laws
are ingrained into every robot's “Positronic Brain” and were
designed to protect a human, and make sure robots were
subservient.
LAW 1:
A robot must never harm a human being, or, through inaction,
allow a human being to come to harm.
LAW 2:
A robot must obey the orders given to it by human beings, except
where such orders would conflict with the First Law.
LAW 3:
A robot must protect its own existence as long as such protection
does not conflict with the First of Second Law.
By implementing these laws, the robots created in his stories, were
a lot safer than they would have been otherwise, and as a result,
could function more efficiently. Asimov's stories do, however,
explore the results of his robots being purely logical machines – at
the end of the day – and taking the orders they are given very
literally.
Towards an Ethical Robot
Steps have been made to attempt to implement something similar
to Asimov's Three Laws into robots to date. Professor Alan Winfield,
Christian Blum, and Wenguo Liu have performed the “Ethical
Robot” experiment. This experiment involves the use of a
“Consequence Engine” that is programmed into one of the robots,
to predict the movements of the other actors (in the case of the
experiment, other robots) and map them on an internal simulation.
It would then simulate various possible movements, and choose
the 'best' solution in order to keep itself, and the other actors safe.
[Figure 1.0] The Ethical Robot Experiment
Extending the Ethical Robot
My report contains the details of expanding the ethical robot's
Consequence Engine from a single 'Smart Robot' unit, to multiple
platforms.
Each robot with the Consequence Engine will be able to simulate
the entire environment, and each actor, meaning they would be
able to predict movements, and, assuming the environment
contains 'DangerZones' they would be able to protect the other
actors in the environment.
The Consequence Engine
As the robot is not designed specifically to save the other actors,
the Consequence Engine is designed as a secondary process that
runs constantly, but doesn't effect the main operation of the robot,
unless necessary. The primary operation of the robots, in these set
of experiments, is to get from A, to B.
As it is navigating the environment to get to the end location, it will
simulate the actions of the other actors, and if one of them was
headed towards a 'DangerZone' then the robot would move to
intercept, altering the actor's course away from the danger.
[Figure 2.0] An Example of the Consequence Engine
The processing of the Consequence Engine predicts multiple
paths in the environment, and then chooses the 'best' of
these. The Red lines in Figure 2.0 are the predicted paths of
the Consequence engine.
The Corridor Experiment
In order for the multiple Consequence Engine code to be
tested, a series of updated experiments needed to be
implemented. Below is a brief overview of the proposed
experiments:
1) The Corridor Experiment: By placing two 'Smart' robots at
opposing ends of a narrow corridor, we can set them going,
and see how they will interact. Their main goal will be to
reach the opposing end of the corridor, and there will not be
any 'dumb' robots to save.
[Figure 3.0] The Corridor Experiment Setup
2) The Ethical Robots: Repeating the Ethical Robot
experiment, but increasing the number of 'Smart' robots,
would mean that the Consequence Engine would be fully
tested in a situation that is likely to commonly arise.
Recommendations for Further Work
Unfortunatly, the scope of my project focused only on the
software side of these experiments. As a result, the Epuck
robots – which were modified with a Linux board extension –
happened to not be powerful enough to run the internal
simulations. The laptop that the code was originally ran on
was not powerful enough to run two instances of this code.
As a result, I have decided to recommend some further work
on the hardware of the Epuck extension board.
These recommendations are as follows:
- Update the Linux extension board, increasing the
processing power available on each Epuck robot, to re-run
this experiment.
- Alternatively, look into alternative methods of processing
the Internal simulations for the robots.
Project Summary
The aim of this project will be to extend existing work on
robots with internal models – i.e. robots with a simulation
of themselves, other robots, and their environment –
inside of themselves. The first step will be to extend an
existing implementation from a single, to multiple robots.
When this has been tested and proven to work, I will then
conduct a series of experiments to show how multiple
robots interacting could behave more safely, or ethically,
than robots without self-simulation.
Project Objectives
• Replicate the 'ethical robot' experiment, and observe
the robots motions, to see where improvement, or
modification can be made.
• Re-write, or modify the code, to allow experimentation
to be taken onto multiple smart robot platforms.
• Run several different experiments, initially in
simulation, then in reality, to a) prove the modified
code runs as desired, and b) to observe the results, and
draw conclusions from there.
Project Conclusion
While it has not been possible to complete the main
experimentation, two of the three main objectives have
been completed. When initially replicating the “Ethical
robot” experiment, we were able to get it running
smoothly, and were able to them use the information
from that to produce the modified code, to be exported
onto each individual platform. What was not expected,
however, was the vast amount of computational power
required to perform this operation – looking back though,
it should have been expected – and the inability to run
multiple instances of the code in parallel, on the current
hardware.
In its current state it has been proven that the
Consequence Engine is capable of:
• Tracking actors in an environment and judging their
'safety' based on preset 'dangerzones'.
• Avoiding Obstacles within the environment, whether
mobile, or static.
• Navigating an environment, and reacting to other
actors' safety, even moving to prevent them from being
in danger.